def detect(save_img=False):
out, source, weights, view_img, save_txt, imgsz = \
opt.output, opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
webcam = source.isnumeric() or source.startswith(
'rtsp') or source.startswith('http') or source.endswith('.txt')
# Initialize
set_logging()
device = select_device(opt.device)
if os.path.exists(out):
shutil.rmtree(out) # delete output folder
os.makedirs(out) # make new output folder
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
if half:
model.half() # to FP16
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(
torch.load('weights/resnet101.pt',
map_location=device)['model']) # load weights
modelc.to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = True
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz)
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)]
for _ in range(len(names))]
# Run inference
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img
) if device.type != 'cpu' else None # run once
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
opt.conf_thres,
opt.iou_thres,
classes=opt.classes,
agnostic=opt.agnostic_nms)
t2 = time_synchronized()
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
print(pred)
# Process detections
for i, det in enumerate(pred): # detections per image
if webcam: # batch_size >= 1
p, s, im0 = path[i], '%g: ' % i, im0s[i].copy()
else:
p, s, im0 = path, '', im0s
save_path = str(Path(out) / Path(p).name)
txt_path = str(Path(out) / Path(p).stem) + (
'_%g' % dataset.frame if dataset.mode == 'video' else '')
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
if det is not None and len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
handles_ymax = []
#handles_xmid = []
handles_ymid = []
handle_mids = []
tailgates_ymin = []
tailgates_ymax = []
tailgate_ythird_coord = []
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += '%g %ss, ' % (n, names[int(c)]) # add to string
# Write results
for *xyxy, conf, cls in reversed(
det): #coords, confidence, classes
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * 5 + '\n') %
(cls, *xywh)) # label format
if save_img or view_img: # Add bbox to image
#label = '%s %.2f' % (names[int(cls)], conf) #confidence not needed
label = '%s ' % (names[int(cls)])
coord1, coord2, dim_label = plot_one_box(
xyxy,
im0,
label=label,
color=colors[int(cls)],
line_thickness=3)
# get important points for line drawing
if int(cls) == 1: #handles
ymax = max(coord1[1], coord2[1])
handles_ymax.append(ymax)
xmid = int((coord1[0] + coord2[0]) / 2)
ymid = int((coord1[1] + coord2[1]) / 2)
handle_mids.append([xmid, ymid])
#cv2.circle(im0, (xmid,ymax), 8, (255,0,0), -1)
elif int(cls) == 0: #tailgates
tailgate_xmin = min(coord1[0], coord2[0])
ymax = max(coord1[1], coord2[1])
tailgates_ymax.append(ymax)
ymin = min(coord1[1], coord2[1])
tailgates_ymin.append(ymin)
tailgate_ythird = int(
abs(coord1[1] - coord2[1]) / 3 + ymin)
tailgate_ythird_coord.append(
[tailgate_xmin, tailgate_ythird])
# added ability to measure between bottom of handle and bottom of tailgate if handle in top 1/3
for i, (handle_mid,
max_point) in enumerate(zip(handle_mids,
handles_ymax)):
hyps = [
hypotenuse(handle_mid, b)
for b in tailgate_ythird_coord
]
closest_index = np.argmin(hyps)
if handle_mid[1] < tailgate_ythird_coord[closest_index][1]:
min_dist_tg = min(
[int(abs(max_point - x)) for x in tailgates_ymax])
start_point = (handle_mid[0], handles_ymax[i])
end_point = (handle_mid[0],
handles_ymax[i] + min_dist_tg)
cv2.line(im0, start_point, end_point, (100, 100, 0), 4)
line_mid = int((start_point[1] + end_point[1]) / 2)
cv2.putText(im0,
label, (start_point[0], line_mid),
0,
1, [0, 0, 0],
thickness=2,
lineType=cv2.LINE_AA)
### Previous ability to measure between bottom of handle and tailgate --- was not robust.
### Keeping until determined not needed
# for i, (mid_point, max_point) in enumerate(zip(handles_ymid, handles_ymax)):
# print(f'\nmidpoint: {mid_point}')
# min_y_dist = min([int(abs(mid_point - x)) for x in tailgate_ythird]) # gets min distance from handle midpoint to tailgate third
# print(f'min y dist: {min_y_dist}')
# print(f'tailgate third: {tailgate_ythird}')
# min_dist_third = min([x for x in tailgate_ythird if abs(x - min_y_dist) in handles_ymid])
# print(f'min_dist_third: {min_dist_third}')
# if mid_point < min_dist_third: #handle mid point in top 1/3 of truck
# min_dist_tg = min([int(abs(max_point - x)) for x in tailgates_ymax])
# print(f'min_dist_tg {min_dist_tg}')
# start_point = (handles_xmid[i], handles_ymax[i])
# print(f'start point: {start_point}')
# end_point = (handles_xmid[i], handles_ymax[i] + min_dist_tg)
# print(f'end point: {end_point}')
# cv2.line(im0, start_point, end_point, (100,100,0), 4)
# label = f'Distance: {min_dist_tg/300:.4f}"L'
# line_mid = int((start_point[1] + end_point[1])/2)
# cv2.putText(im0, label, (start_point[0], line_mid), 0, 1, [0, 0, 0],
# thickness=2, lineType=cv2.LINE_AA)
# Print time (inference + NMS)
print('%sDone. (%.3fs)' % (s, t2 - t1))
# Stream results
if view_img:
cv2.imshow(p, im0)
if cv2.waitKey(1) == ord('q'): # q to quit
raise StopIteration
# Save results (image with detections)
if save_img:
if dataset.mode == 'images':
cv2.imwrite(save_path, im0)
else:
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release(
) # release previous video writer
fourcc = 'mp4v' # output video codec
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
vid_writer = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*fourcc), fps,
(w, h))
vid_writer.write(im0)
if save_txt or save_img:
print('Results saved to %s' % Path(out))
if platform.system() == 'Darwin' and not opt.update: # MacOS
os.system('open ' + save_path)
print('Done. (%.3fs)' % (time.time() - t0))
def run(
weights=ROOT / 'yolov5s.pt', # model.pt path(s)
source=ROOT / 'data/images', # file/dir/URL/glob, 0 for webcam
imgsz=(640, 640), # inference size (height, width)
conf_thres=0.25, # confidence threshold
iou_thres=0.45, # NMS IOU threshold
max_det=1000, # maximum detections per image
device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu
view_img=False, # show results
save_txt=False, # save results to *.txt
save_conf=False, # save confidences in --save-txt labels
save_crop=False, # save cropped prediction boxes
nosave=False, # do not save images/videos
classes=None, # filter by class: --class 0, or --class 0 2 3
agnostic_nms=False, # class-agnostic NMS
augment=False, # augmented inference
visualize=False, # visualize features
update=False, # update all models
project=ROOT / 'runs/detect', # save results to project/name
name='exp', # save results to project/name
exist_ok=False, # existing project/name ok, do not increment
line_thickness=3, # bounding box thickness (pixels)
hide_labels=False, # hide labels
hide_conf=False, # hide confidences
half=False, # use FP16 half-precision inference
dnn=False, # use OpenCV DNN for ONNX inference
):
source = str(source)
save_img = not nosave and not source.endswith(
'.txt') # save inference images
is_file = Path(source).suffix[1:] in (IMG_FORMATS + VID_FORMATS)
is_url = source.lower().startswith(
('rtsp://', 'rtmp://', 'http://', 'https://'))
webcam = source.isnumeric() or source.endswith('.txt') or (is_url
and not is_file)
if is_url and is_file:
source = check_file(source) # download
# Directories
save_dir = increment_path(Path(project) / name,
exist_ok=exist_ok) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(
parents=True, exist_ok=True) # make dir
# Load model
device = select_device(device)
model = DetectMultiBackend(weights, device=device, dnn=dnn)
stride, names, pt, jit, onnx, engine = model.stride, model.names, model.pt, model.jit, model.onnx, model.engine
imgsz = check_img_size(imgsz, s=stride) # check image size
# Half
half &= (
pt or jit or engine
) and device.type != 'cpu' # half precision only supported by PyTorch on CUDA
if pt or jit:
model.model.half() if half else model.model.float()
# Dataloader
if webcam:
view_img = check_imshow()
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz, stride=stride, auto=pt)
bs = len(dataset) # batch_size
else:
dataset = LoadImages(source, img_size=imgsz, stride=stride, auto=pt)
bs = 1 # batch_size
vid_path, vid_writer = [None] * bs, [None] * bs
# Run inference
model.warmup(imgsz=(1, 3, *imgsz), half=half) # warmup
dt, seen = [0.0, 0.0, 0.0], 0
for path, im, im0s, vid_cap, s in dataset:
t1 = time_sync()
im = torch.from_numpy(im).to(device)
im = im.half() if half else im.float() # uint8 to fp16/32
im /= 255 # 0 - 255 to 0.0 - 1.0
if len(im.shape) == 3:
im = im[None] # expand for batch dim
t2 = time_sync()
dt[0] += t2 - t1
# Inference
visualize = increment_path(save_dir / Path(path).stem,
mkdir=True) if visualize else False
pred = model(im, augment=augment, visualize=visualize)
t3 = time_sync()
dt[1] += t3 - t2
# NMS
pred = non_max_suppression(pred,
conf_thres,
iou_thres,
classes,
agnostic_nms,
max_det=max_det)
dt[2] += time_sync() - t3
# Second-stage classifier (optional)
# pred = utils.general.apply_classifier(pred, classifier_model, im, im0s)
# Process predictions
for i, det in enumerate(pred): # per image
seen += 1
if webcam: # batch_size >= 1
p, im0, frame = path[i], im0s[i].copy(), dataset.count
s += f'{i}: '
else:
p, im0, frame = path, im0s.copy(), getattr(dataset, 'frame', 0)
p = Path(p) # to Path
save_path = str(save_dir / p.name) # im.jpg
txt_path = str(save_dir / 'labels' / p.stem) + (
'' if dataset.mode == 'image' else f'_{frame}') # im.txt
s += '%gx%g ' % im.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
imc = im0.copy() if save_crop else im0 # for save_crop
annotator = Annotator(im0,
line_width=line_thickness,
example=str(names))
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(im.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
# Write results
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
line = (cls, *xywh,
conf) if save_conf else (cls,
*xywh) # label format
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
if save_img or save_crop or view_img: # Add bbox to image
c = int(cls) # integer class
label = None if hide_labels else (
names[c]
if hide_conf else f'{names[c]} {conf:.2f}')
annotator.box_label(xyxy, label, color=colors(c, True))
if save_crop:
save_one_box(xyxy,
imc,
file=save_dir / 'crops' / names[c] /
f'{p.stem}.jpg',
BGR=True)
# Print time (inference-only)
LOGGER.info(f'{s}Done. ({t3 - t2:.3f}s)')
# Stream results
im0 = annotator.result()
if view_img:
cv2.imshow(str(p), im0)
cv2.waitKey(1) # 1 millisecond
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
else: # 'video' or 'stream'
if vid_path[i] != save_path: # new video
vid_path[i] = save_path
if isinstance(vid_writer[i], cv2.VideoWriter):
vid_writer[i].release(
) # release previous video writer
if vid_cap: # video
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
else: # stream
fps, w, h = 30, im0.shape[1], im0.shape[0]
save_path += '.mp4'
vid_writer[i] = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps,
(w, h))
vid_writer[i].write(im0)
# Print results
t = tuple(x / seen * 1E3 for x in dt) # speeds per image
LOGGER.info(
f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {(1, 3, *imgsz)}'
% t)
if save_txt or save_img:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")
if update:
strip_optimizer(weights) # update model (to fix SourceChangeWarning)
def detect(save_img=False):
source, weights, view_img, save_txt, imgsz = opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
webcam = source.isnumeric() or source.endswith(
'.txt') or source.lower().startswith(('rtsp://', 'rtmp://', 'http://'))
# Directories
save_dir = Path(
increment_path(Path(opt.project) / opt.name,
exist_ok=opt.exist_ok)) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(
parents=True, exist_ok=True) # make dir
# Initialize
set_logging()
device = select_device(opt.device)
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
if half:
model.half() # to FP16
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(
torch.load('weights/resnet101.pt',
map_location=device)['model']).to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = True
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz)
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in names]
# Run inference
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img
) if device.type != 'cpu' else None # run once
"""
path 图片/视频路径
img 进行resize+pad之后的图片
img0 原size图片
cap 当读取图片时为None，读取视频时为视频源
"""
for path, img, im0s, vid_cap in dataset:
t3 = time_synchronized()
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
"""
前向传播 返回pred的shape是(1, num_boxes, 5+num_class)
h,w为传入网络图片的长和宽，注意dataset在检测时使用了矩形推理，所以这里h不一定等于w
num_boxes = h/32 * w/32 + h/16 * w/16 + h/8 * w/8
pred[..., 0:4]为预测框坐标
预测框坐标为xywh(中心点+宽长)格式
pred[..., 4]为objectness置信度
pred[..., 5:-1]为分类结果
"""
# Inference
t1 = time_synchronized()
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
opt.conf_thres,
opt.iou_thres,
classes=opt.classes,
agnostic=opt.agnostic_nms)
t2 = time_synchronized()
"""
pred:前向传播的输出
conf_thres:置信度阈值
iou_thres:iou阈值
classes:是否只保留特定的类别
agnostic:进行nms是否也去除不同类别之间的框
经过nms之后，预测框格式：xywh-->xyxy(左上角右下角)
pred是一个列表list[torch.tensor]，长度为batch_size
每一个torch.tensor的shape为(num_boxes, 6),内容为box+conf+cls
"""
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
# 对每一张图片作处理
for i, det in enumerate(pred): # detections per image
# 如果输入源是webcam，则batch_size不为1，取出dataset中的一张图片
if webcam: # batch_size >= 1
p, s, im0, frame = path[i], '%g: ' % i, im0s[i].copy(
), dataset.count
else:
p, s, im0, frame = path, '', im0s, getattr(dataset, 'frame', 0)
p = Path(p) # to Path
# 设置保存图片/视频的路径
save_path = str(save_dir / p.name) # img.jpg
# 设置保存框坐标txt文件的路径
txt_path = str(save_dir / 'labels' / p.stem) + (
'' if dataset.mode == 'image' else f'_{frame}') # img.txt
# 设置打印信息(图片长宽)
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
if len(det):
# Rescale boxes from img_size to im0 size
# 调整预测框的坐标：基于resize+pad的图片的坐标-->基于原size图片的坐标
# 此时坐标格式为xyxy
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
# 打印检测到的类别数量
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += f'{n} {names[int(c)]}s, ' # add to string
# Write results
# 保存预测结果
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
# 将xyxy(左上角+右下角)格式转为xywh(中心点+宽长)格式，并除上w，h做归一化，转化为列表再保存
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
line = (cls, *xywh, conf) if opt.save_conf else (
cls, *xywh) # label format
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
# 在原图上画框
if save_img or view_img: # Add bbox to image
label = f'{names[int(cls)]} {conf:.2f}'
plot_one_box(xyxy,
im0,
label=label,
color=colors[int(cls)],
line_thickness=3)
t4 = time_synchronized()
# Print time (inference + NMS)
# 打印前向传播+nms时间
print(f'{s}Done. ({t2 - t1:.3f}s)')
# Stream results
if view_img:
cv2.imshow(str(p), im0)
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
else: # 'video'
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release(
) # release previous video writer
fourcc = 'mp4v' # output video codec
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
vid_writer = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*fourcc), fps,
(w, h))
vid_writer.write(im0)
if save_txt or save_img:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
print(f"Results saved to {save_dir}{s}")
print(f'Done. ({time.time() - t0:.3f}s)')
def detect(opt):
source, weights, view_img, save_txt, imgsz = opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
save_img = not opt.nosave and not source.endswith(
'.txt') # save inference images
webcam = source.isnumeric() or source.endswith(
'.txt') or source.lower().startswith(
('rtsp://', 'rtmp://', 'http://', 'https://'))
# Directories
save_dir = increment_path(Path(opt.project) / opt.name,
exist_ok=opt.exist_ok) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(
parents=True, exist_ok=True) # make dir
# Initialize
set_logging()
device = select_device(opt.device)
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
stride = int(model.stride.max()) # model stride
imgsz = check_img_size(imgsz, s=stride) # check img_size
names = model.module.names if hasattr(
model, 'module') else model.names # get class names
if half:
model.half() # to FP16
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(
torch.load('weights/resnet101.pt',
map_location=device)['model']).to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = check_imshow()
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz, stride=stride)
else:
dataset = LoadImages(source, img_size=imgsz, stride=stride)
# Run inference
if device.type != 'cpu':
model(
torch.zeros(1, 3, imgsz, imgsz).to(device).type_as(
next(model.parameters()))) # run once
t0 = time.time()
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
opt.conf_thres,
opt.iou_thres,
classes=opt.classes,
agnostic=opt.agnostic_nms)
t2 = time_synchronized()
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
if webcam: # batch_size >= 1
#p, s, im0, frame = path[i], f'{i}: ', im0s[i].copy(), dataset.count
p, s, im0, frame = path[i], '', im0s[i].copy(), dataset.count
else:
p, s, im0, frame = path, '', im0s.copy(), getattr(
dataset, 'frame', 0)
p = Path(p) # to Path
save_path = str(save_dir / p.name) # img.jpg
txt_path = str(save_dir / 'labels' / p.stem) + (
'' if dataset.mode == 'image' else f'_{frame}') # img.txt
#s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
imc = im0.copy() if opt.save_crop else im0 # for opt.save_crop
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
# Write results
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
line = (cls, *xywh, conf) if opt.save_conf else (
cls, *xywh) # label format
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
if save_img or opt.save_crop or view_img: # Add bbox to image
c = int(cls) # integer class
label = None if opt.hide_labels else (
names[c]
if opt.hide_conf else f'{names[c]} {conf:.2f}')
plot_one_box(xyxy,
im0,
label=label,
color=colors(c, True),
line_thickness=opt.line_thickness)
if opt.save_crop:
save_one_box(xyxy,
imc,
file=save_dir / 'crops' / names[c] /
f'{p.stem}.jpg',
BGR=True)
# Print time (inference + NMS)
print(f'{s}Done. ({t2 - t1:.3f}s)')
speech(s)
# Stream results
if view_img:
#cv2.imshow(str(p), im0)
cv2.waitKey(1) # 1 millisecond
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
else: # 'video' or 'stream'
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release(
) # release previous video writer
if vid_cap: # video
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
else: # stream
fps, w, h = 30, im0.shape[1], im0.shape[0]
save_path += '.mp4'
vid_writer = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps,
(w, h))
vid_writer.write(im0)
#removing line to reduce delay
#if save_txt or save_img:
#s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
#print(f"Results saved to {save_dir}{s}")
#print(s)
print(f'Done. ({time.time() - t0:.3f}s)')
def detect(source, weights):
result = []
imgsz = 640
dev = ''
# Initialize
set_logging()
device = select_device(dev)
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
stride = int(model.stride.max()) # model stride
imgsz = check_img_size(imgsz, s=stride) # check img_size
if half:
model.half() # to FP16
# Set Dataloader
vid_path, vid_writer = None, None
dataset = LoadImages(source, img_size=imgsz, stride=stride)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in names]
# Run inference
if device.type != 'cpu':
model(torch.zeros(1, 3, imgsz, imgsz).to(device).type_as(next(model.parameters()))) # run once
t0 = time.time()
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=True)[0]
# Apply NMS
pred = non_max_suppression(pred, 0.25, 0.45, classes=0, agnostic=True)
t2 = time_synchronized()
# Process detections
for i, det in enumerate(pred): # detections per image
p, s, im0, frame = path, '', im0s, getattr(dataset, 'frame', 0)
p = Path(p) # to Path
#s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
#s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
# print results
for *xyxy, conf, cls in reversed(det):
nvbb = str(xyxy)
nvbb = nvbb.replace("tensor(", "").replace("[", "").replace("]", "").replace("., device='cuda:0')", "").split(", ")
result.append(nvbb)
return result
def detect():
source, weights, view_img, save_txt, imgsz = opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
state = {0: "red", 1: "green", 2: "yellow", 3: "unknown"}
# Initialize
device = select_device(opt.device)
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
mot_tracker = Sort()
if half:
model.half() # to FP16
# Run inference
t0 = time.time()
dataset = LoadImages(source, img_size=imgsz)
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img) if device.type != 'cpu' else None # run once
traffic_id = 0
data = dict()
frame_id = 0
for path, img, im0s, vid_cap in tqdm(dataset):
data[str(frame_id)] = dict()
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred, opt.conf_thres, opt.iou_thres, classes=opt.classes, agnostic=opt.agnostic_nms)
# Process detections
for i, det in enumerate(pred): # detections per image
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0s.shape).round()
track_bbs_ids = mot_tracker.update(det.cpu()) #[x, y, x, y, id, class]
det = torch.tensor(track_bbs_ids).to(device)
area = (det[:, 0] - det[:, 2]) * (det[:, 1] - det[:, 3])
filter = (area > (im0s.shape[0] * im0s.shape[1]) / 3500)
filter2 = (det[:, 1] < 0.8 * im0s.shape[0])
det = det[filter2]
for i, bbox in enumerate(det):
data[str(frame_id)][str(traffic_id)] = dict()
data[str(frame_id)][str(traffic_id)]["affect"] = bool(filter[i].cpu())
data[str(frame_id)][str(traffic_id)]["coords"] = [int(bbox[0]), int(bbox[1]), int(bbox[2]),
int(bbox[3])]
data[str(frame_id)][str(traffic_id)]["state"] = state[int(bbox[-1])]
traffic_id += 1
else:
track_bbs_ids = mot_tracker.update(np.empty((0, 5)))
frame_id += 1
with open("preds.json", "w") as f:
json.dump(data, f)
print(f'Done. ({time.time() - t0:.3f}s)')
def detect(save_img=False):
out, source, weights, view_img, save_txt, imgsz = \
opt.output, opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
webcam = source.isnumeric() or source.startswith('rtsp') or source.startswith('http') or source.endswith('.txt')
from_yaml = source.endswith('.yaml')
# Initialize
set_logging()
device = select_device(opt.device)
if os.path.exists(out):
shutil.rmtree(out) # delete output folder
os.makedirs(out) # make new output folder
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
if half:
model.half() # to FP16
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(torch.load('weights/resnet101.pt', map_location=device)['model']) # load weights
modelc.to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = True
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz)
elif from_yaml:
save_img = True
with open(source) as f:
data_dict = yaml.load(f, Loader=yaml.FullLoader)
dataset = LoadImages(data_dict['inf'], img_size=imgsz, from_yaml=True)
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in range(len(names))]
# Run inference
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img) if device.type != 'cpu' else None # run once
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred, opt.conf_thres, opt.iou_thres, classes=opt.classes, agnostic=opt.agnostic_nms)
t2 = time_synchronized()
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
if webcam: # batch_size >= 1
p, s, im0 = path[i], '%g: ' % i, im0s[i].copy()
else:
p, s, im0 = path, '', im0s
save_path = str(Path(out) / Path(p).name)
txt_path = str(Path(out) / Path(p).stem) + ('_%g' % dataset.frame if dataset.mode == 'video' else '')
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
if det is not None and len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += '%g %ss, ' % (n, names[int(c)]) # add to string
# Write results
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * 5 + '\n') % (cls, *xywh)) # label format
if save_img or view_img: # Add bbox to image
label = '%s %.2f' % (names[int(cls)], conf)
plot_one_box(xyxy, im0, label=label, color=colors[int(cls)], line_thickness=3)
# Print time (inference + NMS)
print('%sDone. (%.3fs)' % (s, t2 - t1))
# Stream results
if view_img:
cv2.imshow(p, im0)
if cv2.waitKey(1) == ord('q'): # q to quit
raise StopIteration
# Save results (image with detections)
if save_img:
if dataset.mode == 'images':
cv2.imwrite(save_path, im0)
else:
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release() # release previous video writer
fourcc = 'mp4v' # output video codec
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
vid_writer = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*fourcc), fps, (w, h))
vid_writer.write(im0)
if save_txt or save_img:
print('Results saved to %s' % Path(out))
if platform.system() == 'Darwin' and not opt.update: # MacOS
os.system('open ' + save_path)
print('Done. (%.3fs)' % (time.time() - t0))
def predict(weights=weights,
source=source,
save_dir=save_dir,
imgsz=imgsz,
device=device,
conf_thres=conf_thres,
iou_thres=iou_thres,
classes=classes,
augment=True,
agnostic_nms=True,
save_img=False):
# source, weights, view_img, save_txt, imgsz = opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
# webcam = source.isnumeric() or source.endswith('.txt') or source.lower().startswith(
# ('rtsp://', 'rtmp://', 'http://'))
webcam = False
save_txt = False
# Directories
# save_dir = Path(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok)) # increment run
# (save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
# Initialize
set_logging()
device = select_device(device)
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
if half:
model.half() # to FP16
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(
torch.load('weights/resnet101.pt',
map_location=device)['model']).to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = True
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz)
else:
save_img = False
dataset = LoadImages(source, img_size=imgsz)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in names]
# Run inference
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img
) if device.type != 'cpu' else None # run once
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
conf_thres,
iou_thres,
classes=classes,
agnostic=agnostic_nms)
t2 = time_synchronized()
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
if webcam: # batch_size >= 1
p, s, im0, frame = path[i], '%g: ' % i, im0s[i].copy(
), dataset.count
else:
p, s, im0, frame = path, '', im0s, getattr(dataset, 'frame', 0)
p = Path(p) # to Path
# save_path = str(save_dir / p.name) # img.jpg
# txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}') # img.txt
# s += '%gx%g ' % img.shape[2:] # print string
# gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
# for c in det[:, -1].unique():
# n = (det[:, -1] == c).sum() # detections per class
# s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
# Write results
for *xyxy, conf, cls in reversed(det):
#
# if save_txt: # Write to file
# xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
# line = (cls, *xywh, conf) if opt.save_conf else (cls, *xywh) # label format
# with open(txt_path + '.txt', 'a') as f:
# f.write(('%g ' * len(line)).rstrip() % line + '\n')
# if save_img: # Add bbox to image
label = f'{names[int(cls)]} {conf:.2f}'
plot_one_box(xyxy,
im0,
label=label,
color=colors[int(cls)],
line_thickness=3)
# cv2.imwrite(save_dir, im0)
imageb64 = cv2.imencode(".jpg", im0)[1]
base64_data = str(base64.b64encode(imageb64), 'utf-8')
return base64_data
def home_page():
# Nếu là POST (gửi file)
if request.method == "POST":
try:
# Lấy file gửi lên
image = request.files['file']
if image:
# Lưu file
print(image.filename)
print(app.config['UPLOAD_FOLDER'])
source = os.path.join(app.config['UPLOAD_FOLDER'],
image.filename)
print("Save = ", source)
image.save(source)
# source = "data/images/sample4.jpg"
save_img = True
dataset = LoadImages(source, img_size=imgsz, stride=stride)
# Get names and colors
names = model.module.names if hasattr(
model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)]
for _ in names]
# Run inference
if device.type != 'cpu':
model(
torch.zeros(1, 3, imgsz, imgsz).to(device).type_as(
next(model.parameters()))) # run once
conf_thres = 0.25
iou_thres = 0.25
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float(
) # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
pred = model(img, augment=False)[0]
# Apply NMS
pred = non_max_suppression(pred,
conf_thres,
iou_thres,
classes=None,
agnostic=False)
extra = ""
# Process detections
for i, det in enumerate(pred): # detections per image
p, s, im0, frame = path, '', im0s, getattr(
dataset, 'frame', 0)
save_path = source
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:],
det[:, :4],
im0.shape).round()
# Write results
for *xyxy, conf, cls in reversed(det):
if save_img: # Add bbox to image
label = f'{names[int(cls)]} {conf:.2f}'
plot_one_box(xyxy,
im0,
label=label,
color=colors[int(cls)],
line_thickness=3)
extra += "<br>- <b>" + str(names[int(cls)]) + "</b> (" + dict[names[int(cls)]] \
+ ") với độ tin cậy <b>{:.2f}% </b>".format(conf)
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
# Trả về kết quả
return render_template("index.html",
user_image=image.filename,
rand=random.random(),
msg="Tải file lên thành công",
extra=Markup(extra))
else:
# Nếu không có file thì yêu cầu tải file
return render_template('index.html',
msg='Hãy chọn file để tải lên')
except Exception as ex:
# Nếu lỗi thì thông báo
print(ex)
return render_template('index.html',
msg='Không nhận diện được ảnh')
else:
# Nếu là GET thì hiển thị giao diện upload
return render_template('index.html')
def detect(save_txt=False, save_img=True):
img_size = (
320, 192
) if ONNX_EXPORT else opt.img_size # (320, 192) or (416, 256) or (608, 352) for (height, width)
out, source, weights, half, view_img = opt.output, opt.source, opt.weights, opt.half, opt.view_img
webcam = source == '0' or source.startswith('rtsp') or source.startswith(
'http') or source.endswith('.txt')
# Initialize
device = torch_utils.select_device(
device='cpu' if ONNX_EXPORT else opt.device)
def may_mkdirs(d):
if not os.path.exists(d):
os.makedirs(d)
may_mkdirs(out)
img_dir = os.path.join(out, 'images')
may_mkdirs(img_dir)
lbl_dir = os.path.join(out, 'labels')
may_mkdirs(lbl_dir)
# Initialize model
model = Darknet(opt.cfg, img_size)
# Load weights
attempt_download(weights)
if weights.endswith('.pt'): # pytorch format
model.load_state_dict(
torch.load(weights, map_location=device)['model'])
else: # darknet format
_ = load_darknet_weights(model, weights)
# Fuse Conv2d + BatchNorm2d layers
# model.fuse()
# Eval mode
model.to(device).eval()
# Export mode
if ONNX_EXPORT:
img = torch.zeros((1, 3) + img_size) # (1, 3, 320, 192)
torch.onnx.export(model, img, 'weights/export.onnx', verbose=True)
return
# Half precision
half = half and device.type != 'cpu' # half precision only supported on CUDA
if half:
model.half()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = True
torch.backends.cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=img_size, half=half)
else:
save_img = True
dataset = LoadImages(source, img_size=img_size, half=half)
# Get classes and colors
classes = load_classes(parse_data_cfg(opt.data)['names'])
colors = [[random.randint(0, 255) for _ in range(3)]
for _ in range(len(classes))]
# Run inference
frames = 0
for path, img, im0s, vid_cap in dataset:
has_person = False
if frames % 24:
frames += 1
continue
im_name = os.path.splitext(
os.path.basename(path))[0] + f'_{frames}.jpg'
# Get detections
img = torch.from_numpy(img).to(device)
if img.ndimension() == 3:
img = img.unsqueeze(0)
if ONNX_EXPORT:
pred = model(img)
else:
pred, _ = model(img)
if opt.half:
pred = pred.float()
for i, det in enumerate(
non_max_suppression(pred, opt.conf_thres,
opt.nms_thres)): # detections per image
if webcam: # batch_size >= 1
p, s, im0 = path[i], '%g: ' % i, im0s[i]
else:
p, s, im0 = path, '', im0s
save_path = str(Path(out) / Path(p).name)
s += '%gx%g ' % img.shape[2:] # print string
if det is not None and len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += '%g %ss, ' % (n, classes[int(c)]) # add to string
# Write results
h, w, _ = im0.shape
for *xyxy, conf, _, cls in det:
if save_txt: # Write to file
with open(save_path + '.txt', 'a') as file:
file.write(('%g ' * 6 + '\n') % (*xyxy, cls, conf))
if cls == 0:
foo = xyxy2cxcywh(*map(int, xyxy), w, h)
print(0,
*foo,
file=open(
os.path.join(lbl_dir,
im_name.replace('.jpg',
'.txt')), 'a'))
has_person = True
# cv2.drawMarker(im0, (int(foo[0]*w), int(foo[1]*h)), (0,255,0), cv2.MARKER_CROSS)
# if save_img or view_img: # Add bbox to image
# label = '%s %.2f' % (classes[int(cls)], conf)
# plot_one_box(xyxy, im0, label=label, color=colors[int(cls)])
# Stream results
# if view_img:
# cv2.imshow(p, im0)
# Save results (image with detections)
if save_img:
if dataset.mode == 'images':
cv2.imwrite(save_path, im0)
else:
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release(
) # release previous video writer
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
# vid_writer = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*opt.fourcc), fps, (w, h))
# vid_writer.write(im0)
# if has_person:
# cv2.imwrite(os.path.join(img_dir, im_name), im0)
# if cv2.waitKey(1) == 27:
# break
# save image
# save label
frames += 1
if save_txt or save_img:
print('Results saved to %s' % os.getcwd() + os.sep + out)
if platform == 'darwin': # MacOS
os.system('open ' + out + ' ' + save_path)
def detect(save_img=False):
#--- MANU READ Model Background
cv2.namedWindow("Detection", cv2.WINDOW_NORMAL)
cv2.namedWindow("Remove", cv2.WINDOW_NORMAL)
#backgroundModel = cv2.imread("/home/manuel/visiont3lab-github/public/ai_library/storage/images/backgroundModelRome.png",1)
#backgroundModel = cv2.imread("/home/manuel/visiont3lab-github/public/ai_library/storage/images/backgroundModelTokyo.png",1)
#backgroundModel = cv2.imread("/home/manuel/visiont3lab-github/public/ai_library/storage/images/backgroundModelCastenaso.png",1)
#backgroundModel = cv2.imread("/home/manuel/visiont3lab-github/public/ai_library/storage/images/backgroundModelLidlImola.png",1)
backgroundModel = cv2.imread("/home/manuel/visiont3lab-github/public/ai_library/storage/images/backgroundModelCMB.png",1)
eB = EstimateBackground(backgroundModel.shape)
clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))
#cv2.imshow("jk",backgroundModel)
#cv2.waitKey(0)
# ------
source, weights, view_img, save_txt, imgsz = opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
webcam = source.isnumeric() or source.endswith('.txt') or source.lower().startswith(
('rtsp://', 'rtmp://', 'http://'))
# Directories
save_dir = Path(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok)) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
# Initialize
set_logging()
device = select_device(opt.device)
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
stride = int(model.stride.max()) # model stride
imgsz = check_img_size(imgsz, s=stride) # check img_size
if half:
model.half() # to FP16
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(torch.load('weights/resnet101.pt', map_location=device)['model']).to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = True
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz, stride=stride)
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz, stride=stride)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in names]
# Run inference
if device.type != 'cpu':
model(torch.zeros(1, 3, imgsz, imgsz).to(device).type_as(next(model.parameters()))) # run once
t0 = time.time()
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=opt.augment)[0]
print("------HERE")
print(img.shape, opt.augment)
# Apply NMS
pred = non_max_suppression(pred, opt.conf_thres, opt.iou_thres, classes=opt.classes, agnostic=opt.agnostic_nms)
t2 = time_synchronized()
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
if webcam: # batch_size >= 1
p, s, im0, frame = path[i], '%g: ' % i, im0s[i].copy(), dataset.count
else:
p, s, im0, frame = path, '', im0s, getattr(dataset, 'frame', 0)
# --- MANU
#cv2.imwrite("results/im.png",im0)
detectionMask = np.zeros(backgroundModel.shape,np.uint8)
imgMask = im0.copy()
imgDraw = im0.copy()
# -----
p = Path(p) # to Path
save_path = str(save_dir / p.name) # img.jpg
txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}') # img.txt
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
# Write results
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
line = (cls, *xywh, conf) if opt.save_conf else (cls, *xywh) # label format
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
if save_img or view_img: # Add bbox to image
label = f'{names[int(cls)]} {conf:.2f}'
# --- MANU
imgMask,detectionMask,coord = plot_one_box_custom(xyxy, imgMask, detectionMask,label=label)
# ---
imgDraw = plot_one_box(xyxy, imgDraw, label=label, color=colors[int(cls)], line_thickness=1)
# Print time (inference + NMS)
print(f'{s}Done. ({t2 - t1:.3f}s)')
# Stream results
if view_img:
#cv2.imshow(str(p), im0 )
#cv2.imshow("backgroundModelInit", backgroundModel)
#backgroundModel_hsv = cv2.cvtColor(backgroundModel, cv2.COLOR_RGB2HSV)
#im0_hsv = cv2.cvtColor(im0, cv2.COLOR_RGB2HSV)
#cv2.imshow("b",backgroundModel_hsv[:,:,2])
#cv2.imshow("i",im0_hsv[:,:,2])
#a = match_histograms(backgroundModel_hsv[:,:,2],im0_hsv[:,:,2], multichannel=False)
#cv2.imshow("br",np.uint8(a))
#cv2.waitKey(0)
#backgroundModel = cv2.cvtColor(backgroundModel_hsv, cv2.COLOR_HSV2RGB)
#cv2.imshow("backgroundModelEnd", backgroundModel)
#cv2.waitKey(0)
#detectionMaskGray = cv2.cvtColor(detectionMask,cv2.COLOR_RGB2GRAY)
#inpaint = cv2.inpaint(imgMask,detectionMaskGray,2,cv2.INPAINT_TELEA)
#cv2.imshow("inpainting", inpaint)
#cv2.imshow("background", backgroundModel)
#backgroundModel = match_histograms(backgroundModel,inpaint, multichannel=True)
#backgroundModel_hsv = cv2.cvtColor(backgroundModel, cv2.COLOR_RGB2HSV)
#inpaint_hsv = cv2.cvtColor(inpaint, cv2.COLOR_RGB2HSV)
#backgroundModel_hsv[:,:,2] = match_histograms(backgroundModel_hsv[:,:,2], inpaint_hsv[:,:,2], multichannel=False)
#backgroundModel_hsv[:,:,2] = backgroundModel_hsv[:,:,2] - 40
#backgroundModel = cv2.cvtColor(backgroundModel_hsv, cv2.COLOR_HSV2RGB)
'''
imgs,coords = splitImages(inpaint)
for c in coords:
i,j,w,h = c
backgroundModel_hsv = cv2.cvtColor(backgroundModel[i:i+w,j:j+h], cv2.COLOR_RGB2HSV)
im0_hsv = cv2.cvtColor(inpaint[i:i+w,j:j+h], cv2.COLOR_RGB2HSV)
backgroundModel_hsv[:,:,2] = match_histograms(backgroundModel_hsv[:,:,2],im0_hsv[:,:,2], multichannel=False)
backgroundModel[i:i+w,j:j+h] = cv2.cvtColor(backgroundModel_hsv, cv2.COLOR_HSV2RGB)
#backgroundModel[i:i+w,j:j+h] = match_histograms(backgroundModel[i:i+w,j:j+h], im0[i:i+w,j:j+h], multichannel=True)
#cv2.imshow("img",img)
#cv2.waitKey(0)
'''
# Remove People
#maskDetectionInv = cv2.bitwise_not(maskDetection)
#backgroundModel = match_histograms(backgroundModel, im0, multichannel=True)
#backgroundModel = eB.updated(im0)
backgroundModelMask = cv2.bitwise_and(backgroundModel,detectionMask)
now = datetime.now()
now = now.strftime("%d_%H_%M_%S_%f")
cv2.imshow("Detection", imgDraw)
cv2.imshow("BackgroundModel", backgroundModel)
cv2.imshow("Image", imgMask)
#img_hsv = cv2.cvtColor(res, cv2.COLOR_RGB2HSV)
#img_hsv[:, :, 2] = cv2.equalizeHist(img_hsv[:, :, 2])
#img_hsv[:, :, 2] = clahe.apply(img_hsv[:, :, 2])
#res = cv2.cvtColor(img_hsv, cv2.COLOR_HSV2RGB)
res = cv2.addWeighted(imgMask, 1, backgroundModelMask, 1, 0)
cv2.imwrite("/home/manuel/visiont3lab-github/public/people-remove/images/results/Molino-Ariani-yolov5/"+now+"_yolo.png",imgDraw)
cv2.imwrite("/home/manuel/visiont3lab-github/public/people-remove/images/results/Molino-Ariani-yolov5/"+now+"_inapaint.png",res)
#res_hsv = cv2.cvtColor(res, cv2.COLOR_RGB2HSV)
#res_hsv[:, :, 2] = cv2.equalizeHist(res_hsv[:, :, 2])
#res_hsv[:,:,0] = clahe.apply(res_hsv[:,:,0])
#res = cv2.cvtColor(res_hsv, cv2.COLOR_HSV2RGB)
#backgroundModel[i:i+w,j:j+h] = match_histograms(backgroundModel[i:i+w,j:j+h], im0[i:i+w,j:j+h], multichannel=True)
cv2.imshow("Remove",res)
cv2.waitKey(1)
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
else: # 'video'
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release() # release previous video writer
fourcc = 'mp4v' # output video codec
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
vid_writer = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*fourcc), fps, (w, h))
vid_writer.write(im0)
if save_txt or save_img:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
print(f"Results saved to {save_dir}{s}")
print(f'Done. ({time.time() - t0:.3f}s)')
def AL_detect(opt):
# Detect ảnh
# File weight model, nguồn dữ liệu detect, Kích cỡ ảnh sử dụng
weights, source, imgsz = opt.weights, opt.source, opt.img_size
# khởi tạo
device = select_device(opt.device)
half = device.type != 'cpu'
# Load model
model = attempt_load(weights=weights, map_location=device)
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
if half:
model.half() # to FP16
dataset = LoadImages(source, img_size=imgsz)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
# Run inference
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img
) if device.type != 'cpu' else None # run once
result = {}
# duyệt tất cả các ảnh
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
# chuuẩn hóa hình ảnh
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
opt.conf_thres,
opt.iou_thres,
classes=opt.classes,
agnostic=opt.agnostic_nms)
t2 = time_synchronized()
# Process detections
result[path] = []
for i, det in enumerate(pred): # detections per image
p, s, im0, frame = path, '', im0s, getattr(dataset, 'frame', 0)
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += f'{n} {names[int(c)]}s, ' # add to string
# Lưu thông tin về box vào 1 file
for *xyxy, conf, cls in reversed(det):
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh)
x, y, w, h = xywh
data = {
"class": cls.item(),
"box": [x, y, w, h],
"conf": conf.item()
}
result[path].append(data)
# Print time (inference + NMS)
print(f'{s}Done. ({t2 - t1:.3f}s)')
print(f'Done. ({time.time() - t0:.3f}s)')
return result
def detect(weights='mdp/weights/weights.pt',
source_address='http://192.168.15.1:8008',
img_size=416,
conf_thres=0.01,
iou_thres=0.5,
device='',
classes=None,
agnostic_nms=False,
augment=False,
update=False,
scale_percent=50):
source = source_address + '/stream.mjpg'
# tcp server
HOST = ''
PORT = 8080
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
print('Socket created')
s.bind((HOST, PORT))
print('Socket bind complete')
s.listen(10)
print('Socket now listening')
conn, addr = s.accept()
predicted_label = None
imgsz = img_size
webcam = source.isnumeric() or source.startswith(
'rtsp') or source.startswith('http') or source.endswith('.txt')
# Initialize
set_logging()
device = select_device(device)
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
if half:
model.half() # to FP16
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(
torch.load('weights/resnet101.pt',
map_location=device)['model']) # load weights
modelc.to(device).eval()
# Set Dataloader
if webcam:
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz)
else:
dataset = LoadImages(source, img_size=imgsz)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)]
for _ in range(len(names))]
# Run inference
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img
) if device.type != 'cpu' else None # run once
row_num = 0
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
pred = model(img, augment=augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
conf_thres,
iou_thres,
classes=classes,
agnostic=agnostic_nms)
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
if webcam: # batch_size >= 1
p, s, im0 = path[i], '%g: ' % i, im0s[i].copy()
else:
p, s, im0 = path, '', im0s
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
if det is not None and len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].detach().unique():
n = (det[:, -1] == c).sum() # detections per class
s += '%g %ss, ' % (n, names[int(c)]) # add to string
# Write results
for *xyxy, conf, cls in det:
predicted_label = names[int(cls)]
if predicted_label:
if not image_seen[predicted_label]:
label_id = label_id_mapping.get(predicted_label)
if conf < confidence_threshold(
label_id
): # fine tune for up arrow (white)
break
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
good, text = check_bounding_box(
xywh, im0.shape[0], im0.shape[1])
if not good:
break
print(('%s ' * 5 + '\n') %
(label_id, *xywh)) # label format
image_seen[predicted_label] = True
# determine image position
x, y, w, h = xywh
conn.sendall(
bytes(
json.dumps({
'label': label_id,
'x': x,
'y': y
}), 'utf-8')) # send result to algo
label = '%s %.2f' % (label_id, conf)
plot_one_box(xyxy,
im0,
label=label,
color=colors[int(cls)],
line_thickness=3)
#percent by which the image is resized
# scale_percent = 50
#calculate the 50 percent of original dimensions
width = int(im0.shape[1] * scale_percent / 100)
height = int(im0.shape[0] * scale_percent / 100)
# dsize
dsize = (width, height)
# resize image
im0 = cv2.resize(im0, dsize)
# detected_images.append(im0)
row_num = append_image(im0, row_num)
def vconcat_resize_min(
im_list, interpolation=cv2.INTER_CUBIC):
w_min = min(im.shape[1] for im in im_list)
im_list_resize = [
cv2.resize(im, (w_min,
int(im.shape[0] * w_min /
im.shape[1])),
interpolation=interpolation)
for im in im_list
]
return cv2.vconcat(im_list_resize)
def hconcat_resize_min(
im_list, interpolation=cv2.INTER_CUBIC):
h_min = min(im.shape[0] for im in im_list)
im_list_resize = [
cv2.resize(im, (int(im.shape[1] * h_min /
im.shape[0]), h_min),
interpolation=interpolation)
for im in im_list
]
return cv2.hconcat(im_list_resize)
def concat_tile_resize(
im_list_2d, interpolation=cv2.INTER_CUBIC):
im_list_v = [
hconcat_resize_min(
im_list_h,
interpolation=cv2.INTER_CUBIC)
for im_list_h in im_list_2d
]
return vconcat_resize_min(
im_list_v, interpolation=cv2.INTER_CUBIC)
im_tile = concat_tile_resize(detected_images)
cv2.imshow('ImageWindow', im_tile)
cv2.imwrite('result.png', im_tile)
break
if cv2.waitKey(1) == ord('q'): # q to quit
raise StopIteration
def detect(save_img=False):
out, source, weights, view_img, save_txt, imgsz = \
opt.output, opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
webcam = source.isnumeric() or source.startswith(
'rtsp') or source.startswith('http') or source.endswith('.txt')
# Initialize
set_logging()
device = select_device(opt.device)
if os.path.exists(out):
shutil.rmtree(out) # delete output folder
os.makedirs(out) # make new output folder
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
if half:
model.half() # to FP16
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(
torch.load('weights/resnet101.pt',
map_location=device)['model']) # load weights
modelc.to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = True
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz)
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)]
for _ in range(len(names))]
# Find index corresponding to a person
idx_person = names.index("person")
# Deep SORT: initialize the tracker
cfg = get_config()
cfg.merge_from_file(opt.config_deepsort)
deepsort = DeepSort(cfg.DEEPSORT.REID_CKPT,
max_dist=cfg.DEEPSORT.MAX_DIST,
min_confidence=cfg.DEEPSORT.MIN_CONFIDENCE,
nms_max_overlap=cfg.DEEPSORT.NMS_MAX_OVERLAP,
max_iou_distance=cfg.DEEPSORT.MAX_IOU_DISTANCE,
max_age=cfg.DEEPSORT.MAX_AGE,
n_init=cfg.DEEPSORT.N_INIT,
nn_budget=cfg.DEEPSORT.NN_BUDGET,
use_cuda=True)
# AlphaPose: initialization
args_p = update_config(opt.config_alphapose)
cfg_p = update_config(args_p.ALPHAPOSE.cfg)
args_p.ALPHAPOSE.tracking = args_p.ALPHAPOSE.pose_track or args_p.ALPHAPOSE.pose_flow
demo = SingleImageAlphaPose(args_p.ALPHAPOSE, cfg_p, device)
output_pose = opt.output.split('/')[0] + '/pose'
if not os.path.exists(output_pose):
os.mkdir(output_pose)
# Run inference
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img
) if device.type != 'cpu' else None # run once
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
opt.conf_thres,
opt.iou_thres,
classes=opt.classes,
agnostic=opt.agnostic_nms)
t2 = time_synchronized()
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
if webcam: # batch_size >= 1
p, s, im0 = path[i], '%g: ' % i, im0s[i].copy()
else:
p, s, im0 = path, '', im0s
save_path = str(Path(out) / Path(p).name)
txt_path = str(Path(out) / Path(p).stem) + (
'_%g' % dataset.frame if dataset.mode == 'video' else '')
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
if det is not None and len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += '%g %ss, ' % (n, names[int(c)]) # add to string
# Deep SORT: person class only
idxs_ppl = (
det[:, -1] == idx_person
).nonzero(as_tuple=False).squeeze(
dim=1) # 1. List of indices with 'person' class detections
dets_ppl = det[idxs_ppl, :
-1] # 2. Torch.tensor with 'person' detections
print('\n {} people were detected!'.format(len(idxs_ppl)))
# Deep SORT: convert data into a proper format
xywhs = xyxy2xywh(dets_ppl[:, :-1]).to("cpu")
confs = dets_ppl[:, 4].to("cpu")
# Deep SORT: feed detections to the tracker
if len(dets_ppl) != 0:
trackers, features = deepsort.update(xywhs, confs, im0)
for i, d in enumerate(trackers):
plot_one_box(d[:-1],
im0,
label='ID' + str(int(d[-1])),
color=colors[1],
line_thickness=1)
# AlphaPose: prepare YOLOv5 outputs in alphapose format and find pose heat maps as well as skeleton key points
if len(trackers) > 0:
pose = demo.process('frame_' + str(dataset.frame), im0,
trackers)
# visualization
im0 = demo.vis(im0, pose)
# write the result to json:
demo.writeJson([pose],
output_pose,
form=args_p.ALPHAPOSE.format,
for_eval=args_p.ALPHAPOSE.eval)
# Print time (inference + NMS)
print('%sDone. (%.3fs)' % (s, t2 - t1))
# Stream results
if view_img:
cv2.imshow(p, im0)
if cv2.waitKey(1) == ord('q'): # q to quit
raise StopIteration
# Save results (image with detections)
if save_img:
if dataset.mode == 'images':
cv2.imwrite(save_path, im0)
else:
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release(
) # release previous video writer
fourcc = 'mp4v' # output video codec
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
vid_writer = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*fourcc), fps,
(w, h))
vid_writer.write(im0)
if save_txt or save_img:
print('Results saved to %s' % Path(out))
if platform.system() == 'Darwin' and not opt.update: # MacOS
os.system('open ' + save_path)
print('Done. (%.3fs)' % (time.time() - t0))
def detect(image, weights):
# Final Declaration
final_det = None
final_class = []
final_text = []
source, weights, save_txt, imgsz, ocr, save_img = image, weights, True, 1280, True, True
webcam = source.isnumeric() or source.endswith(
'.txt') or source.lower().startswith(('rtsp://', 'rtmp://', 'http://'))
# Directories
save_dir = Path(increment_path(Path('runs/detect') / 'exp',
exist_ok=False)) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(
parents=True, exist_ok=True) # make dir
# Initialize
set_logging()
device = select_device('')
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
stride = int(model.stride.max()) # model stride
imgsz = check_img_size(imgsz, s=stride) # check img_size
if half:
model.half() # to FP16
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(
torch.load('weights/resnet101.pt',
map_location=device)['model']).to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = check_imshow()
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz, stride=stride)
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz, stride=stride)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in names]
# Run inference
if device.type != 'cpu':
model(
torch.zeros(1, 3, imgsz, imgsz).to(device).type_as(
next(model.parameters()))) # run once
t0 = time.time()
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=False)[0]
# Apply NMS
pred = non_max_suppression(pred,
0.6,
0.45,
classes=None,
agnostic=False)
t2 = time_synchronized()
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
if webcam: # batch_size >= 1
p, s, im0, frame = path[i], '%g: ' % i, im0s[i].copy(
), dataset.count
else:
p, s, im0, frame = path, '', im0s, getattr(dataset, 'frame', 0)
p = Path(p) # to Path
save_path = str(save_dir / p.name) # img.jpg
txt_path = str(save_dir / 'labels' / p.stem) + (
'' if dataset.mode == 'image' else f'_{frame}') # img.txt
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
if len(det):
# Rescale boxes from img_size to im0 size
print()
print(img.shape[2:], " ---------------------- ", im0.shape)
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
final_det = det[:, :4].tolist()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
# python detect.py --source test\ --weights weights\model.pt --save-txt --save-conf
final_class_reverse = []
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
line = (conf, cls,
*xywh) if True else (cls,
*xywh) # label format
# print(names[int(cls)], str(line[0].squeeze().tolist()), 'box(xywh) = ' + str(line[2:]))
final_class_reverse.append(
names[int(cls)] + '(' +
str(round((line[0].squeeze().tolist()), 2)) + ')')
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
if save_img or view_img: # Add bbox to image
label = f'{names[int(cls)]} {conf:.2f}'
plot_one_box(xyxy,
im0,
label=label,
color=colors[int(cls)],
line_thickness=2)
for i in final_class_reverse:
final_class.append(i)
# Print time (inference + NMS)
print(f'{s}Done. ({t2 - t1:.3f}s)')
if ocr:
# Opens a image in RGB mode
im = Image.open(path)
newlist = []
# pytesseract.pytesseract.tesseract_cmd = "/app/.apt/usr/bin/tesseract"
for I in final_det:
# [27.0, 14.0, 177.0, 91.0]
left, top, right, bottom = I[0], I[1], I[2], I[3]
im1 = im.crop((left, top, right, bottom))
text = pytesseract.image_to_string(im1)
# print(text.split('\n'))
newlist = text.split('\n')
newlist.pop(-1)
final_text.append(' '.join(map(str, newlist)))
# im1.show()
print('LENGHTS', len(final_class), len(final_text))
final_class.reverse()
res = dict(zip(final_class, final_text))
print("Resultant dictionary is : " + str(res))
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
else: # 'video'
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release(
) # release previous video writer
fourcc = 'mp4v' # output video codec
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
vid_writer = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*fourcc), fps,
(w, h))
vid_writer.write(im0)
if save_txt or save_img:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
print(f"Results saved to {save_dir}{s}")
return res
def detect(save_img=False):
save_obj = True
source, weights, view_img, save_txt, imgsz = opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
webcam = source.isnumeric() or source.endswith(
'.txt') or source.lower().startswith(('rtsp://', 'rtmp://', 'http://'))
# Directories
save_dir = Path(
increment_path(Path(opt.project) / opt.name,
exist_ok=opt.exist_ok)) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(
parents=True, exist_ok=True) # make dir
# Initialize
set_logging()
device = select_device(opt.device)
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
stride = int(model.stride.max()) # model stride
imgsz = check_img_size(imgsz, s=stride) # check img_size
if half:
model.half() # to FP16
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(
torch.load('weights/resnet101.pt',
map_location=device)['model']).to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = check_imshow()
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz, stride=stride)
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz, stride=stride)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in names]
# Run inference
if device.type != 'cpu':
model(
torch.zeros(1, 3, imgsz, imgsz).to(device).type_as(
next(model.parameters()))) # run once
t0 = time.time()
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
opt.conf_thres,
opt.iou_thres,
classes=opt.classes,
agnostic=opt.agnostic_nms)
t2 = time_synchronized()
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
if webcam: # batch_size >= 1
p, s, im0, frame = path[i], '%g: ' % i, im0s[i].copy(
), dataset.count
else:
p, s, im0, frame = path, '', im0s, getattr(dataset, 'frame', 0)
p = Path(p) # to Path
save_path = str(save_dir / p.name) # img.jpg
txt_path = str(save_dir / 'labels' / p.stem) + (
'' if dataset.mode == 'image' else f'_{frame}') # img.txt
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
# Write results
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
line = (cls, *xywh, conf) if opt.save_conf else (
cls, *xywh) # label format
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
if save_img or view_img: # Add bbox to image
label = f'{names[int(cls)]} {conf:.2f}'
plot_one_box(xyxy,
im0,
label=label,
color=colors[int(cls)],
line_thickness=3)
if save_obj:
for k in range(len(det)):
x, y, w, h = int(xyxy[0]), int(
xyxy[1]), int(xyxy[2] - xyxy[0]), int(xyxy[3] -
xyxy[1])
img_ = im0.astype(np.uint8)
crop_img = img_[y:y + h, x:x + w]
#!!rescale image !!!
filename = label + '{:}.jpg'.format(+1)
filepath = os.path.join(
r'./inference/detected_obj/', filename)
cv2.imwrite(filepath, crop_img)
#saving results as csv!!!
'''crop_img = Image.open(crop_img)
# get original image parameters...
width, height = crop_img.size
format = crop_img.format
mode = crop_img.mode
#make the gray scale
img_grey = img_crop.convert('L')
value = np.asarray(img_grey.getdata(), dtype=np.int).reshape((img_grey.size[1], img_grey.size[0]))'''
value = crop_img.flatten()
print(value)
with open(
"/content/drive/MyDrive/yolov5/inference/csv/images.csv",
'a') as f:
writer = csv.writer(f)
writer.writerow(value)
else:
print("There is no detected object")
continue
# Print time (inference + NMS)
print(f'{s}Done. ({t2 - t1:.3f}s)')
# Stream results
if view_img:
cv2.imshow(str(p), im0)
cv2.waitKey(1) # 1 millisecond
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
else: # 'video'
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release(
) # release previous video writer
fourcc = 'mp4v' # output video codec
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
vid_writer = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*fourcc), fps,
(w, h))
vid_writer.write(im0)
if save_txt or save_img:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
print(f"Results saved to {save_dir}{s}")
print(f'Done. ({time.time() - t0:.3f}s)')
def detect(save_img=False):
out, source, weights, view_img, save_txt, imgsz = \
opt.output, opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
webcam = source == '0' or source.startswith('rtsp') or source.startswith(
'http') or source.endswith('.txt')
# initialize deepsort
cfg = get_config()
cfg.merge_from_file(opt.config_deepsort)
deepsort = DeepSort(cfg.DEEPSORT.REID_CKPT,
max_dist=cfg.DEEPSORT.MAX_DIST,
min_confidence=cfg.DEEPSORT.MIN_CONFIDENCE,
nms_max_overlap=cfg.DEEPSORT.NMS_MAX_OVERLAP,
max_iou_distance=cfg.DEEPSORT.MAX_IOU_DISTANCE,
max_age=cfg.DEEPSORT.MAX_AGE,
n_init=cfg.DEEPSORT.N_INIT,
nn_budget=cfg.DEEPSORT.NN_BUDGET,
use_cuda=True)
# Initialize
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
if os.path.exists(out):
shutil.rmtree(out) # delete output folder
if not os.path.exists(opt.smooth_txt):
os.makedirs(opt.smooth_txt)
os.makedirs(out) # make new output folder
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
if half:
model.half() # to FP16
# # Second-stage classifier
# classify = False
# if classify:
# modelc = load_classifier(name='resnet101', n=2) # initialize
# modelc.load_state_dict(torch.load('weights/resnet101.pt', map_location=device)['model']) # load weights
# modelc.to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = True
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz)
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[np.random.randint(0, 255) for _ in range(3)]
for _ in range(len(names))]
# Run inference
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img
) if device.type != 'cpu' else None # run once
# dataset contains all the frames (or images) of the video
crds_crop = np.empty(
(0, 4)) # contains coordinates of a single bbox with the highest conf
np_nan = np.empty([1, 4]) # for tracking
np_nan[:] = np.nan # for tracking
frame_no = 0
for path, img, im0s, vid_cap in dataset: # im0s, img - initial, resized and padded (img)
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
opt.conf_thres,
opt.iou_thres,
classes=opt.classes,
agnostic=opt.agnostic_nms)
t2 = time_synchronized()
# # Apply Classifier
# if classify:
# pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
if webcam: # batch_size >= 1
p, s, im0 = path[i], '%g: ' % i, im0s[i].copy()
else:
p, s, im0 = path, '', im0s
save_path = str(Path(out) / Path(p).name)
txt_path = str(Path(out) / Path(p).stem) + (
'_%g' % dataset.frame if dataset.mode == 'video' else '')
s += '%gx%g ' % img.shape[2:] # print string
# gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
if det is not None and len(det): # only when obj is in the frame
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# these will be used for the deepsort input
xywhs = xyxy2xywh(det[:, :4].cpu())
confs = det[:, 4].cpu()
# Pass detections to deepsort
outputs = deepsort.update(xywhs, confs,
im0) # this is numpy array
###########################################################
# FOR NOW, WE WILL ONLY BE KEEPING THE MOST CONFIDENT VALUE
###########################################################
max_conf_id = confs.argmax()
# keeping the coordinates row with max conf (det now only keeps one row and four columns)
det = det[max_conf_id, :].reshape(1, 6)
to_append = xyxy2xywh(det[:, :4].cpu().numpy().reshape(
1, 4).astype(int))
if len(crds_crop) == 0:
crds_crop = np.append(crds_crop, to_append).reshape(-1, 4)
else:
crds_crop = np.append(crds_crop, to_append, axis=0)
# draw boxes for visualization
if len(outputs) > 0:
bbox_xyxy = outputs[:, :4]
identities = outputs[:, -1]
# draw_boxes(im0, bbox_xyxy, identities) # no tracking boxes for now
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += '%g %ss, ' % (n, names[int(c)]) # add to string
# Write results
for *xyxy, conf, cls in det:
if save_txt: # Write to file
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * 5 + '\n') %
(*xyxy, conf)) # label format
if save_img or view_img: # Add bbox to image
label = '%s' % (names[int(cls)])
plot_one_box(xyxy,
im0,
label=label,
color=colors[int(cls)],
line_thickness=2)
else:
deepsort.increment_ages()
if len(crds_crop) == 0:
crds_crop = np.append(crds_crop, np_nan).reshape(-1, 4)
else:
crds_crop = np.append(crds_crop, np_nan, axis=0)
# Print time (inference + NMS)
print('%sDone. (%.3fs)' % (s, t2 - t1))
# Stream results
if view_img:
cv2.imshow(p, im0)
if cv2.waitKey(1) == ord('q'): # q to quit
raise StopIteration
# Save results (image with detections)
if save_img:
if dataset.mode == 'images':
cv2.imwrite(save_path, im0)
else:
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release(
) # release previous video writer
fourcc = 'mp4v' # output video codec
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
vid_writer = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*fourcc), fps,
(w, h))
vid_writer.write(im0)
frame_no += 1
##############################################################################
# this part should be temporary as online filtering will be implemented
crds_crop = linear_interp(crds_crop)
max_side_bbox = crds_crop[:, 2:].max(axis=1) * 1.2 # 20% relaxation
# making sure that the window size does not exceed frame size
max_side_bbox = np.where(max_side_bbox < min(w, h), max_side_bbox,
min(w, h))
crds_crop = np.c_[crds_crop, max_side_bbox]
crds_crop = smoothing(crds_crop, fps)
np.savetxt(os.path.join(opt.smooth_txt,
os.path.basename(path)[:-4] + '_savgol_' + '.txt'),
crds_crop,
delimiter=' ')
if save_txt or save_img:
print('Results saved to %s' % Path(out))
if platform == 'darwin' and not opt.update: # MacOS
os.system('open ' + save_path)
print('Done. (%.3fs)' % (time.time() - t0))
def detect(save_img=False):
source, weights, weights2, view_img, save_txt, imgsz, thres = opt.source, opt.weights, opt.weights2, opt.view_img, opt.save_txt, opt.img_size, opt.headThres
save_img = not opt.nosave and not source.endswith(
'.txt') # save inference images
webcam = source.isnumeric() or source.endswith(
'.txt') or source.lower().startswith(
('rtsp://', 'rtmp://', 'http://', 'https://'))
# Directories
save_dir = Path(opt.project)
# save_dir = Path(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok)) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(
parents=True, exist_ok=True) # make dir
# Initialize
set_logging()
device = select_device(opt.device)
half = device.type != 'cpu' # half precision only supported on CUDA
# Load models
model1 = attempt_load(weights, map_location=device)
model2 = attempt_load(weights2, map_location=device)
stride1 = int(model1.stride.max()) # model strides
stride2 = int(model2.stride.max()) # model 2 strides
names1 = model1.module.names if hasattr(model1, 'module') else model1.names
names2 = model2.module.names if hasattr(model2, 'module') else model2.names
imgsz = check_img_size(imgsz, s=stride1) # check img_size
if half:
model1.half() # to FP16
model2.half() # to FP16 too
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(
torch.load('weights/resnet101.pt',
map_location=device)['model']).to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
if opt.saveWebcam:
save_img = True
view_img = check_imshow()
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz, stride=stride1)
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz, stride=stride1)
# Run inference
numFrames = 1
t0 = time.time()
numWeapons = 0
headDet = []
weapDet = []
frames = []
mask = None
for path, img, im0s, vid_cap in dataset:
print("\nFrame:", numFrames)
if webcam:
print("FPS", dataset.fps)
t1 = time_synchronized()
# Adding to frame
if (len(img.shape) >= 4):
myImg = np.dstack((img[0, 0], img[0, 1], img[0, 2]))
else:
myImg = np.dstack((img[0], img[1], img[2]))
# Creating mask
if mask is None:
mask = np.zeros_like(myImg)
# Appending frames
frames.append(myImg)
if len(frames) > opt.filterLen:
frames.pop(0)
# Starting with the actual detections
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Do first round of predictions
model = model1 # set pointer to model1
names = names1
# Inference
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
opt.conf_thres,
opt.iou_thres,
opt.classes,
opt.agnostic_nms,
max_det=opt.max_det)
t2 = time_synchronized()
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
weapDet.append(det.clone())
if len(weapDet) > opt.filterLen:
weapDet.pop(0)
if webcam: # batch_size >= 1
p, s, im0, frame = path[i], '%g: ' % i, im0s[i], dataset.count
else:
p, s, im0, frame = path, '', im0s, getattr(dataset, 'frame', 0)
p = Path(p) # to Path
save_path = str(save_dir / p.name) # img.jpg
txt_path = str(save_dir / 'labels' / p.stem) + (
'' if dataset.mode == 'image' else f'_{frame}') # img.txt
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
imc = im0.copy() if opt.save_crop else im0 # for opt.save_crop
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# for detection in det[:, :4]:
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += f"{n} {names2[int(c)]}{'s' * (n > 1)}, " # add to string
# Write results
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
line = (cls, *xywh, conf) if opt.save_conf else (
cls, *xywh) # label format
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
if save_img or view_img: # Add bbox to image
c = int(cls) # integer class
label = None if opt.hide_labels else (
names[c]
if opt.hide_conf else f'{names[c]} {conf:.2f}')
plot_one_box(xyxy,
im0,
label=label,
color=colors(c, True),
line_thickness=opt.line_thickness)
if opt.save_crop:
save_one_box(xyxy,
imc,
file=Path(save_path + '_crops') /
names[c] / f'{p.stem}.jpg',
BGR=True)
print("2nd Round")
model = model2
names = names2
# Inference
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
opt.conf_thres,
opt.iou_thres,
opt.classes,
opt.agnostic_nms,
max_det=opt.max_det)
t2 = time_synchronized()
# Process detections
for i, det in enumerate(pred): # detections per image
numWeapons += len(det)
headDet.append(det.clone())
if len(headDet) > opt.filterLen:
headDet.pop(0)
if webcam: # batch_size >= 1
p, s, im0, frame = path[i], '%g: ' % i, im0s[i], dataset.count
else:
p, s, im0, frame = path, '', im0s, getattr(dataset, 'frame', 0)
p = Path(p) # to Path
save_path = str(save_dir / p.name) # img.jpg
txt_path = str(save_dir / 'labels' / p.stem) + (
'' if dataset.mode == 'image' else f'_{frame}') # img.txt
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
imc = im0.copy() if opt.save_crop else im0 # for opt.save_crop
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
# Write results
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
line = (cls, *xywh, conf) if opt.save_conf else (
cls, *xywh) # label format
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
if save_img or view_img: # Add bbox to image
c = int(cls) # integer class
label = None if opt.hide_labels else (
names[c]
if opt.hide_conf else f'{names[c]} {conf:.2f}')
width = round(
float((xyxy[2] - xyxy[0]) / im0.shape[1]), 2)
plot_one_box(xyxy,
im0,
label=label + " " + str(width),
color=colors(c, True),
line_thickness=opt.line_thickness)
# Print time (inference + NMS)
print(f'{s}Done. ({t2 - t1:.3f}s)')
# Checking for Spider-Sense
sense = spider_sense(headDet, weapDet, frames, im0, thres, mask,
device)
if sense[0] or sense[1]:
cv2.putText(im0, "Spider-Sense Tingling!", (30, 90),
cv2.FONT_HERSHEY_SIMPLEX, 3, (255, 0, 0), 5)
# Stream results
if view_img:
if opt.flowShow:
thisMask = cv2.resize(mask, (im0.shape[1], im0.shape[0]))
im0 = cv2.add(im0, thisMask)
cv2.imshow(str(p), im0)
cv2.waitKey(1) # 1 millisecond
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
elif webcam:
if vid_path != save_path + ".mp4":
vid_path = save_path + ".mp4"
print("Save Path: ", save_path)
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release(
) # release previous video writer
fourcc = 'mp4v'
fps = dataset.fps
w = dataset.w
h = dataset.h
vid_writer = cv2.VideoWriter(
vid_path, cv2.VideoWriter_fourcc(*fourcc), fps,
(w, h))
if opt.flowShow:
thisMask = cv2.resize(mask,
(im0.shape[1], im0.shape[0]))
im0 = cv2.add(im0, thisMask)
vid_writer.write(im0)
else: # 'video'
if vid_path != save_path: # new video
vid_path = save_path
print("Save Path: ", save_path)
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release(
) # release previous video writer
fourcc = 'mp4v' # output video codec
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
vid_writer = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*fourcc), fps,
(w, h))
if opt.flowShow:
thisMask = cv2.resize(mask,
(im0.shape[1], im0.shape[0]))
im0 = cv2.add(im0, thisMask)
vid_writer.write(im0)
# Checking break condition
if numFrames == opt.maxFrames:
break
numFrames += 1
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release() # release previous video writer
if save_txt or save_img:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
def detect(save_img=False):
out, source, weights, view_img, save_txt, imgsz = \
opt.output, opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
webcam = source.isnumeric() or source.startswith(('rtsp://', 'rtmp://', 'http://')) or source.endswith('.txt')
# Initialize
set_logging()
device = select_device(opt.device)
if os.path.exists(out):
shutil.rmtree(out) # delete output folder
os.makedirs(out) # make new output folder
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
if half:
model.half() # to FP16
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = True
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz)
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in range(len(names))]
# Run inference
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img) if device.type != 'cpu' else None # run once
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred, opt.conf_thres, opt.iou_thres, classes=opt.classes, agnostic=opt.agnostic_nms)
t2 = time_synchronized()
# Process detections
for i, det in enumerate(pred): # detections per image
p, s, im0 = path, '', im0s
save_path = str(Path(out) / Path(p).name)
txt_path = str(Path(out) / Path(p).stem) + ('_%g' % dataset.frame if dataset.mode == 'video' else '')
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
if det is not None and len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round()
max =0
maxlabel=""
xyxymax=[]
for *xyxy, conf, cls in reversed(det): #가장큰 차량 찾기 (주차하는 차량 찾기)
if( names[int(cls)] == 'car' or names[int(cls)] == 'truck' or names[int(cls)] == 'bus'):
if((int(xyxy[0])-int(xyxy[1]))**2 + (int(xyxy[2])-int(xyxy[3])**2)**-2) > max:
max = ((int(xyxy[0])-int(xyxy[1]))**2 + (int(xyxy[2])-int(xyxy[3])**2)**-2)
maxlabel = '%s %.2f' % (names[int(cls)], conf)
xyxymax = xyxy[0],xyxy[1],xyxy[2],xyxy[3]
crop_image = im0[int(xyxy[1]):int(xyxy[3]) ,int(xyxy[0]):int(xyxy[2])]
# Write results
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * 5 + '\n') % (cls, *xywh)) # label format
if names[int(cls)] == 'bus' or names[int(cls)] == 'car' or names[int(cls)] == 'truck':
pass
if save_img or view_img: # Add bbox to image
label = '%s %.2f' % (names[int(cls)], conf)
plot_one_box(xyxy, im0, label=label, color=colors[int(cls)], line_thickness=3)
# Stream results
if False:
cv2.imshow(p, im0)
if cv2.waitKey(1) == ord('q'): # q to quit
raise StopIteration
def detect(self, _img, save_img=False):
model = self.model
save_txt = False
save_img = True
imgsz = 640
conf_thres = 0.25
iou_thres = 0.45
classes = None
agnostic_nms = False
view_img = False
# Initialize
set_logging()
device = ''
device = select_device(device)
half = device.type != 'cpu' # half precision only supported on CUDA
bbox_list = []
label_list = []
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(
torch.load('weights/resnet101.pt',
map_location=device)['model']).to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
dataset = LoadImages(_img, img_size=imgsz)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in names]
# Run inference
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img
) if device.type != 'cpu' else None # run once
for img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=False)[0]
# Apply NMS
pred = non_max_suppression(pred,
conf_thres,
iou_thres,
classes=classes,
agnostic=agnostic_nms)
t2 = time_synchronized()
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
s, im0 = '', im0s
#save_path = str(save_dir / p.name)
#txt_path = str(save_dir / 'labels' / p.stem) + ('_%g' % dataset.frame if dataset.mode == 'video' else '')
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += '%g %ss, ' % (n, names[int(c)]) # add to string
# Write results
for *xyxy, conf, cls in reversed(det):
#if save_img or view_img: # Add bbox to image
label = '%s %.2f' % (names[int(cls)], conf)
plot_one_box(xyxy,
im0,
label=label,
color=colors[int(cls)],
line_thickness=3)
for item in xyxy:
#print(item)
bbox_list.append(item.cpu().numpy())
label_list.append(label)
bbox_array = np.array(bbox_list)
bbox_array = bbox_array.reshape(-1, 4)
# Print time (inference + NMS)
#print('%sDone. (%.3fs)' % (s, t2 - t1))
# Stream results
if view_img:
cv2.imshow("QQ", im0)
if cv2.waitKey(0) == ord('q'): # q to quit
raise StopIteration
# Save results (image with detections)
confidence = []
label_name = []
for label in label_list:
split_label = label.split(' ')
confidence.append(float(split_label[-1]))
label_name.append(split_label[0])
return im0, bbox_array, confidence, label_name
def run(
weights='yolov5s.pt', # model.pt path(s)
source='data/images', # file/dir/URL/glob, 0 for webcam
imgsz=640, # inference size (pixels)
conf_thres=0.25, # confidence threshold
iou_thres=0.45, # NMS IOU threshold
max_det=1000, # maximum detections per image
device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu
view_img=False, # show results
save_txt=False, # save results to *.txt
save_conf=False, # save confidences in --save-txt labels
save_crop=False, # save cropped prediction boxes
nosave=False, # do not save images/videos
classes=None, # filter by class: --class 0, or --class 0 2 3
agnostic_nms=False, # class-agnostic NMS
augment=False, # augmented inference
visualize=False, # visualize features
update=False, # update all models
project='runs/detect', # save results to project/name
name='exp', # save results to project/name
exist_ok=False, # existing project/name ok, do not increment
line_thickness=3, # bounding box thickness (pixels)
hide_labels=False, # hide labels
hide_conf=False, # hide confidences
half=False, # use FP16 half-precision inference
):
save_img = not nosave and not source.endswith(
'.txt') # save inference images
webcam = source.isnumeric() or source.endswith(
'.txt') or source.lower().startswith(
('rtsp://', 'rtmp://', 'http://', 'https://'))
# Directories
save_dir = increment_path(Path(project) / name,
exist_ok=exist_ok) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(
parents=True, exist_ok=True) # make dir
# Initialize
set_logging()
device = select_device(device)
half &= device.type != 'cpu' # half precision only supported on CUDA
# Load model
w = weights[0] if isinstance(weights, list) else weights
classify, suffix, suffixes = False, Path(w).suffix.lower(), [
'.pt', '.onnx', '.tflite', '.pb', ''
]
check_suffix(w, suffixes) # check weights have acceptable suffix
pt, onnx, tflite, pb, saved_model = (suffix == x
for x in suffixes) # backend booleans
stride, names = 64, [f'class{i}' for i in range(1000)] # assign defaults
if pt:
model = attempt_load(weights, map_location=device) # load FP32 model
stride = int(model.stride.max()) # model stride
names = model.module.names if hasattr(
model, 'module') else model.names # get class names
if half:
model.half() # to FP16
if classify: # second-stage classifier
modelc = load_classifier(name='resnet50', n=2) # initialize
modelc.load_state_dict(
torch.load('resnet50.pt',
map_location=device)['model']).to(device).eval()
elif onnx:
check_requirements(('onnx', 'onnxruntime'))
import onnxruntime
session = onnxruntime.InferenceSession(w, None)
else: # TensorFlow models
check_requirements(('tensorflow>=2.4.1', ))
import tensorflow as tf
if pb: # https://www.tensorflow.org/guide/migrate#a_graphpb_or_graphpbtxt
def wrap_frozen_graph(gd, inputs, outputs):
x = tf.compat.v1.wrap_function(
lambda: tf.compat.v1.import_graph_def(gd, name=""),
[]) # wrapped import
return x.prune(
tf.nest.map_structure(x.graph.as_graph_element, inputs),
tf.nest.map_structure(x.graph.as_graph_element, outputs))
graph_def = tf.Graph().as_graph_def()
graph_def.ParseFromString(open(w, 'rb').read())
frozen_func = wrap_frozen_graph(gd=graph_def,
inputs="x:0",
outputs="Identity:0")
elif saved_model:
model = tf.keras.models.load_model(w)
elif tflite:
interpreter = tf.lite.Interpreter(
model_path=w) # load TFLite model
interpreter.allocate_tensors() # allocate
input_details = interpreter.get_input_details() # inputs
output_details = interpreter.get_output_details() # outputs
int8 = input_details[0][
'dtype'] == np.uint8 # is TFLite quantized uint8 model
imgsz = check_img_size(imgsz, s=stride) # check image size
ascii = is_ascii(names) # names are ascii (use PIL for UTF-8)
# Dataloader
if webcam:
view_img = check_imshow()
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz, stride=stride, auto=pt)
bs = len(dataset) # batch_size
else:
dataset = LoadImages(source, img_size=imgsz, stride=stride, auto=pt)
bs = 1 # batch_size
vid_path, vid_writer = [None] * bs, [None] * bs
# Run inference
if pt and device.type != 'cpu':
model(
torch.zeros(1, 3, *imgsz).to(device).type_as(
next(model.parameters()))) # run once
dt, seen = [0.0, 0.0, 0.0], 0
for path, img, im0s, vid_cap in dataset:
t1 = time_sync()
if onnx:
img = img.astype('float32')
else:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img = img / 255.0 # 0 - 255 to 0.0 - 1.0
if len(img.shape) == 3:
img = img[None] # expand for batch dim
t2 = time_sync()
dt[0] += t2 - t1
# Inference
if pt:
visualize = increment_path(save_dir / Path(path).stem,
mkdir=True) if visualize else False
pred = model(img, augment=augment, visualize=visualize)[0]
elif onnx:
pred = torch.tensor(
session.run([session.get_outputs()[0].name],
{session.get_inputs()[0].name: img}))
else: # tensorflow model (tflite, pb, saved_model)
imn = img.permute(0, 2, 3, 1).cpu().numpy() # image in numpy
if pb:
pred = frozen_func(x=tf.constant(imn)).numpy()
elif saved_model:
pred = model(imn, training=False).numpy()
elif tflite:
if int8:
scale, zero_point = input_details[0]['quantization']
imn = (imn / scale + zero_point).astype(
np.uint8) # de-scale
interpreter.set_tensor(input_details[0]['index'], imn)
interpreter.invoke()
pred = interpreter.get_tensor(output_details[0]['index'])
if int8:
scale, zero_point = output_details[0]['quantization']
pred = (pred.astype(np.float32) -
zero_point) * scale # re-scale
pred[..., 0] *= imgsz[1] # x
pred[..., 1] *= imgsz[0] # y
pred[..., 2] *= imgsz[1] # w
pred[..., 3] *= imgsz[0] # h
pred = torch.tensor(pred)
t3 = time_sync()
dt[1] += t3 - t2
# NMS
pred = non_max_suppression(pred,
conf_thres,
iou_thres,
classes,
agnostic_nms,
max_det=max_det)
dt[2] += time_sync() - t3
# Second-stage classifier (optional)
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process predictions
for i, det in enumerate(pred): # per image
seen += 1
if webcam: # batch_size >= 1
p, s, im0, frame = path[i], f'{i}: ', im0s[i].copy(
), dataset.count
else:
p, s, im0, frame = path, '', im0s.copy(), getattr(
dataset, 'frame', 0)
p = Path(p) # to Path
save_path = str(save_dir / p.name) # img.jpg
txt_path = str(save_dir / 'labels' / p.stem) + (
'' if dataset.mode == 'image' else f'_{frame}') # img.txt
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
imc = im0.copy() if save_crop else im0 # for save_crop
annotator = Annotator(im0,
line_width=line_thickness,
pil=not ascii)
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
# Write results
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
line = (cls, *xywh,
conf) if save_conf else (cls,
*xywh) # label format
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
if save_img or save_crop or view_img: # Add bbox to image
c = int(cls) # integer class
label = None if hide_labels else (
names[c]
if hide_conf else f'{names[c]} {conf:.2f}')
annotator.box_label(xyxy, label, color=colors(c, True))
if save_crop:
save_one_box(xyxy,
imc,
file=save_dir / 'crops' / names[c] /
f'{p.stem}.jpg',
BGR=True)
# Print time (inference-only)
print(f'{s}Done. ({t3 - t2:.3f}s)')
# Stream results
im0 = annotator.result()
if view_img:
cv2.imshow(str(p), im0)
cv2.waitKey(1) # 1 millisecond
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
else: # 'video' or 'stream'
if vid_path[i] != save_path: # new video
vid_path[i] = save_path
if isinstance(vid_writer[i], cv2.VideoWriter):
vid_writer[i].release(
) # release previous video writer
if vid_cap: # video
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
else: # stream
fps, w, h = 30, im0.shape[1], im0.shape[0]
save_path += '.mp4'
vid_writer[i] = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps,
(w, h))
vid_writer[i].write(im0)
# Print results
t = tuple(x / seen * 1E3 for x in dt) # speeds per image
print(
f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {(1, 3, *imgsz)}'
% t)
if save_txt or save_img:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
print(f"Results saved to {colorstr('bold', save_dir)}{s}")
if update:
strip_optimizer(weights) # update model (to fix SourceChangeWarning)
def detect(self, data):
"""
返回的bboxes是实际的坐标，x1，y1，x2，y2，是左上角和右下角的坐标
:param data: 图片数据的列表 [image1, image2]
:return: [[images, bboxes, confidences, labels],[images, bboxes,confidences, labels],...] bboxes是所有的bboxes, confidence是置信度， labels是所有的bboxes对应的label，
"""
#检查设置的图片的大小和模型的步长是否能整除
imgsz = check_img_size(self.img_size, s=self.stride) # check img_size
#下载数据集, images保存图片的本地的路径
images = []
for url in data:
if url.startswith('http'):
image = self.download_file(url, self.image_dir)
else:
#不是http的格式的，是本地文件的，那么直接使用即可
image = url
images.append(image)
#设置数据集
dataset = LoadImages(path=self.image_dir,
img_size=imgsz,
stride=self.stride)
# 这里我们重设下images，我们只要自己需要的images既可, dataset.nf, 即number_files, 文件数量也需要修改下
dataset.files = images
dataset.nf = len(images)
dataset.video_flag = [False] * len(images)
# 设置模型
predict_model = self.predict_model
# 运行推理
if self.device.type != 'cpu':
predict_model(
torch.zeros(1, 3, imgsz, imgsz).to(self.device).type_as(
next(predict_model.parameters()))) # run once
#计算耗时
t0 = time.time()
# path是图片的路径，img是图片的改变size后的numpy格式[channel, new_height,new_witdh], im0s是原始的图片,[height, width,channel], eg: (2200, 1700, 3), vid_cap 是None如果是图片，只对视频有作用
results = []
for idx, (path, img, im0s, vid_cap) in enumerate(dataset):
# 如果是GPU，会放到GPU上
img = torch.from_numpy(img).to(self.device)
#转换成float
img = img.float()
#归一化
img /= 255.0 # 0 - 255 to 0.0 - 1.0
#扩展一个batch_size维度, [batch_isze, channel, new_height, new_witdh], eg: torch.Size([1, 3, 640, 512])
if img.ndimension() == 3:
img = img.unsqueeze(0)
#开始推理, time_synchronized是GPU的同步
t1 = time_synchronized()
# pred模型的预测结果 [batch_size,hidden_size, other] eg: torch.Size([1, 20160, 8]), 8代表 (x1, y1, x2, y2, conf, cls1, cls2, cls3...), 前4个是bbox坐标，conf是置信度，cls是类别的，cls1代表是类别1的概率
pred = predict_model(img, augment=False)[0]
#使用 NMS, pred 是一个列表，里面是,一个元素代表一张图片, 一个元素的维度是 [bbox_num, other], other代表(x1, y1, x2, y2, conf, max_cls_prob) eg: torch.Size([5, 6])
pred = non_max_suppression(pred,
self.conf_thres,
self.iou_thres,
classes=self.classes,
agnostic=self.agnostic_nms)
t2 = time_synchronized()
# 处理 detections,i是索引，det是所有的bbox, torch.Size([3, 6])，代表3个bbox，6代表 (x1, y1, x2, y2, 置信度, 类别id)
for i, det in enumerate(pred): # detections per image
# s 是初始化一个空字符串，用于打印预测结果，im0是原始图片, frame是对于视频而言的, 这里pred一定是以个元素，因为我们迭代的是一张图片
p, s, im0, frame = path, '', im0s, getattr(dataset, 'frame', 0)
#p原始图片的绝对路径
p = Path(p) # to Path
save_path = os.path.join(self.predict_dir,
'pt_' + p.name) #预测后的保存的图片的路径
s += ' 预测的图片尺寸高度宽度%gx%g，' % img.shape[
2:] # print string, eg '640x480 '
s += '原始尺寸为高度宽度%sx%s，' % im0.shape[:
2] # print string, eg '640x480 '
# 图片的width,height, width, height, eg: tensor([1700, 2200, 1700, 2200]), 用于下面的归一化
#如果det不为空，说明检测到了bbox，检测到了目标
if len(det):
# bbox 放大到原始图像的大小，从img_size 到原始图像 im0 尺寸， bbox左上角的x1，y1, 右下角的x2,y2
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# eg: [[832.0, 160.0, 1495.0, 610.0, 0.9033942818641663], [849.0, 1918.0, 1467.0, 2016.0, 0.8640206456184387], [204.0, 0.0, 798.0, 142.0, 0.2842876613140106]]
bboxes = det[:, :4].tolist()
confidences = det[:, 4].tolist()
# eg: ['figure', 'equation', 'figure']
labels = [
self.label_list_cn[i]
for i in map(int, det[:, -1].tolist())
]
#图片的名称，bboex，置信度，标签，都加到结果, 原始图像的尺寸
results.append([
images[idx], bboxes, confidences, labels, im0.shape[:2]
])
#最后一个维度的最后一位是预测的结果, unique是为了统计多个相同的结果
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += f"{n}个{self.label_list[int(c)]}{'s' * (n > 1)} bbox, " # eg: 2个figures bbox
# Write results
for *xyxy, conf, cls in reversed(det):
if self.save_img or self.view_img: # 给图片添加bbox，画到图片上, 坐标也画到图上
a, b, c, d = int(xyxy[0]), int(xyxy[1]), int(
xyxy[2]), int(xyxy[3])
label = f'{self.label_list[int(cls)]} {conf:.2f},{a}:{b}:{c}:{d}'
plot_one_box(xyxy,
im0,
label=label,
color=self.colors[int(cls)],
line_thickness=3)
# Print time (inference + NMS)
print(f'{s}完成. 耗时({t2 - t1:.3f}s)')
else:
#如果没有匹配到，那么为空
results.append([images[idx], [], [], [], []])
print(f'{s}完成. 没有发现目标,耗时({t2 - t1:.3f}s)')
# 保存图片的识别结果到目录下
if self.save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
print(f"保存识别结果到 {save_path}{s}")
print(f'Done. ({time.time() - t0:.3f}s)')
return results
def detect(save_img=False):
source, weights, view_img, save_txt, imgsz = opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
webcam = source.isnumeric() or source.endswith(
'.txt') or source.lower().startswith(('rtsp://', 'rtmp://', 'http://'))
# Directories
save_dir = Path(
increment_path(Path(opt.project) / opt.name,
exist_ok=opt.exist_ok)) # increment run
(save_dir / 'labels' if save_txt else save_dir).mkdir(
parents=True, exist_ok=True) # make dir
# Initialize
set_logging()
device = select_device(opt.device)
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
if half:
model.half() # to FP16
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(
torch.load('weights/resnet101.pt',
map_location=device)['model']).to(device).eval()
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = True
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz)
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in names]
# Run inference
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img
) if device.type != 'cpu' else None # run once
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
opt.conf_thres,
opt.iou_thres,
classes=opt.classes,
agnostic=opt.agnostic_nms)
t2 = time_synchronized()
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
if webcam: # batch_size >= 1
p, s, im0, frame = Path(
path[i]), '%g: ' % i, im0s[i].copy(), dataset.count
else:
p, s, im0, frame = Path(path), '', im0s, getattr(
dataset, 'frame', 0)
save_path = str(save_dir / p.name)
txt_path = str(save_dir / 'labels' / p.stem) + (
'' if dataset.mode == 'image' else f'_{frame}')
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += f'{n} {names[int(c)]}s, ' # add to string
# Write results
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
line = (cls, *xywh, conf) if opt.save_conf else (
cls, *xywh) # label format
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * len(line)).rstrip() % line + '\n')
if save_img or view_img: # Add bbox to image
label = f'{names[int(cls)]} {conf:.2f}'
plot_one_box(xyxy,
im0,
label=label,
color=colors[int(cls)],
line_thickness=3)
# Print time (inference + NMS)
print(f'{s}Done. ({t2 - t1:.3f}s)')
# Stream results
if view_img:
cv2.imshow(str(p), im0)
if cv2.waitKey(1) == ord('q'): # q to quit
raise StopIteration
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite(save_path, im0)
else: # 'video'
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release(
) # release previous video writer
fourcc = 'mp4v' # output video codec
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
vid_writer = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*fourcc), fps,
(w, h))
vid_writer.write(im0)
if save_txt or save_img:
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
print(f"Results saved to {save_dir}{s}")
print(f'Done. ({time.time() - t0:.3f}s)')
def detect(q1,
q2,
weights='yolov5s.pt',
source='inference/images',
conf=0.4,
output='inference/output',
img=640,
iou=0.5,
device='',
view='store_true',
save='store_true',
classes='+',
agnostic='store_true',
augment='store_true',
update='store_true',
save_img=False):
parser = argparse.ArgumentParser()
parser.add_argument('--weights',
nargs='+',
type=str,
default=f'{weights}',
help='model.pt path(s)')
parser.add_argument('--source',
type=str,
default=f'{source}',
help='source') # file/folder, 0 for webcam
parser.add_argument('--output',
type=str,
default=f'{output}',
help='output folder') # output folder
parser.add_argument('--img-size',
type=int,
default=img,
help='inference size (pixels)')
parser.add_argument('--conf-thres',
type=float,
default=conf,
help='object confidence threshold')
parser.add_argument('--iou-thres',
type=float,
default=iou,
help='IOU threshold for NMS')
parser.add_argument('--device',
default=f'{device}',
help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--view-img', action=f'{view}', help='display results')
parser.add_argument('--save-txt',
action=f'{save}',
help='save results to *.txt')
parser.add_argument('--classes',
nargs=f'{classes}',
type=int,
help='filter by class: --class 0, or --class 0 2 3')
parser.add_argument('--agnostic-nms',
action=f'{agnostic}',
help='class-agnostic NMS')
parser.add_argument('--augment',
action=f'{augment}',
help='augmented inference')
parser.add_argument('--update',
action=f'{update}',
help='update all models')
opt = parser.parse_args()
out, source, weights, view_img, save_txt, imgsz = \
opt.output, opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
# Initialize
set_logging()
device = select_device(opt.device)
if os.path.exists(out):
shutil.rmtree(out) # delete output folder
os.makedirs(out) # make new output folder
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
model = attempt_load(weights, map_location=device) # load FP32 model
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
if half:
model.half() # to FP16
# Run inference
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img
) if device.type != 'cpu' else None # run once
# Second-stage classifier
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(
torch.load('weights/resnet101.pt',
map_location=device)['model']) # load weights
modelc.to(device).eval()
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)]
for _ in range(len(names))]
source = opt.source
webcam = source.isnumeric() or source.startswith(
('rtsp://', 'rtmp://', 'http://')) or source.endswith('.txt')
# Set Dataloader
vid_path, vid_writer = None, None
if webcam:
view_img = True
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadStreams(source, img_size=imgsz)
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz)
s = []
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = time_synchronized()
pred = model(img, augment=opt.augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
opt.conf_thres,
opt.iou_thres,
classes=opt.classes,
agnostic=opt.agnostic_nms)
t2 = time_synchronized()
#여기까지 타임 재는 곳
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process detections
for i, det in enumerate(pred): # detections per image
if webcam: # batch_size >= 1
p, im0 = path[i], im0s[i].copy()
else:
p, im0 = path, im0s
save_path = str(Path(out) / Path(p).name)
txt_path = str(Path(out) / Path(p).stem) + (
'_%g' % dataset.frame if dataset.mode == 'video' else '')
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
if det is not None and len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
if names[int(c)] not in s:
s.append(names[int(c)]) # add to string
# Write results
for *xyxy, conf, cls in reversed(det):
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * 5 + '\n') %
(cls, *xywh)) # label format
if save_img or view_img: # Add bbox to image
label = '%s %.2f' % (names[int(cls)], conf)
plot_one_box(xyxy,
im0,
label=label,
color=colors[int(cls)],
line_thickness=3)
# Print time (inference + NMS)
print('%sDone. (%.3fs)' % (s, t2 - t1))
# Save results (imqage with detections)
if save_img:
if dataset.mode == 'images':
cv2.imwrite(save_path, im0)
else:
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release(
) # release previous video writer
fourcc = 'mp4v' # output video codec
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
vid_writer = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*fourcc), fps,
(w, h))
vid_writer.write(im0)
q1.put(" ".join(s))
q2.put(im0)
if save_txt or save_img:
print('Results savqed to %s' % Path(out))