def display(self, pprint=False, show=False, save=False, crop=False, render=False, save_dir=Path('')):
for i, (im, pred) in enumerate(zip(self.imgs, self.pred)):
str = f'image {i + 1}/{len(self.pred)}: {im.shape[0]}x{im.shape[1]} '
if pred is not None:
for c in pred[:, -1].unique():
n = (pred[:, -1] == c).sum() # detections per class
str += f"{n} {self.names[int(c)]}{'s' * (n > 1)}, " # add to string
if show or save or render or crop:
for *box, conf, cls in pred: # xyxy, confidence, class
label = f'{self.names[int(cls)]} {conf:.2f}'
if crop:
save_one_box(box, im, file=save_dir / 'crops' / self.names[int(cls)] / self.files[i])
else: # all others
plot_one_box(box, im, label=label, color=colors(cls))
im = Image.fromarray(im.astype(np.uint8)) if isinstance(im, np.ndarray) else im # from np
if pprint:
print(str.rstrip(', '))
if show:
im.show(self.files[i]) # show
if save:
f = self.files[i]
im.save(save_dir / f) # save
print(f"{'Saved' * (i == 0)} {f}", end=',' if i < self.n - 1 else f' to {save_dir}\n')
if render:
self.imgs[i] = np.asarray(im)
def display(self, pprint=False, show=False, save=False, crop=False, render=False, save_dir=Path('')):
for i, (im, pred) in enumerate(zip(self.imgs, self.pred)):
str = f'image {i + 1}/{len(self.pred)}: {im.shape[0]}x{im.shape[1]} '
if pred.shape[0]:
for c in pred[:, -1].unique():
n = (pred[:, -1] == c).sum() # detections per class
str += f"{n} {self.names[int(c)]}{'s' * (n > 1)}, " # add to string
if show or save or render or crop:
annotator = Annotator(im, pil=not self.ascii)
for *box, conf, cls in reversed(pred): # xyxy, confidence, class
label = f'{self.names[int(cls)]} {conf:.2f}'
if crop:
save_one_box(box, im, file=save_dir / 'crops' / self.names[int(cls)] / self.files[i])
else: # all others
annotator.box_label(box, label, color=colors(cls))
im = annotator.im
else:
str += '(no detections)'
im = Image.fromarray(im.astype(np.uint8)) if isinstance(im, np.ndarray) else im # from np
if pprint:
LOGGER.info(str.rstrip(', '))
if show:
im.show(self.files[i]) # show
if save:
f = self.files[i]
im.save(save_dir / f) # save
if i == self.n - 1:
LOGGER.info(f"Saved {self.n} image{'s' * (self.n > 1)} to {colorstr('bold', save_dir)}")
if render:
self.imgs[i] = np.asarray(im)
def plot(self, im, det):
'''names = self.names; colors = self.colors
for *xyxy, conf, c in reversed(det): # (x1,y1,x2,y2,conf,cls)
c = int(c); label = f'{names[c]} {conf:.2f}' # Add bbox to image
plot_one_box(xyxy, im, label=label, color=colors[c], line_width=2)#'''
names = self.names
from utils.plots import colors
annotator = Annotator(im, line_width=2, example=str(names))
for *xyxy, conf, c in reversed(det): # (x1,y1,x2,y2,conf,cls)
c = int(c)
label = f'{names[c]} {conf:.2f}' # Add bbox to image
annotator.box_label(xyxy, label, color=colors(c, True))
im[:] = annotator.result() #'''
return {
names[int(c)]: int((det[:, -1] == c).sum())
for c in det[:, -1].unique()
}
def __call__(self, image, *args, **kwargs):
img = self.image_preprocess(image)
""" Inference """
pred = self.model(img)[0]
""" Apply NMS """
det = non_max_suppression(pred, self.conf_thres, self.iou_thres)[0]
""" Process detections """
im0 = image.copy()
s = ''
bbox_container = []
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 """
""" detections per class """
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum()
""" add to string """
s += f"{n} {self.names[int(c)]}{'s' * (n > 1)}, "
""" Write results """
for *xyxy, conf, cls in reversed(det):
c = int(cls)
label = f'{self.names[c]} {conf:.2f}'
""" xyxy: LU --> RD """
plot_one_box(xyxy,
im0,
label=label,
color=colors(c, True),
line_thickness=2)
bbox = {
'class': self.names[c],
'confidence': round(conf.item(), 2),
'box': [int(v.item()) for v in xyxy]
}
bbox_container.append(bbox)
print(s)
return bbox_container
def detect_and_annotate(self, img_in):
detections, img_out = self.get_detections(img_in)
# Loop over detections
first = self.names.copy() if self.names else None
for xywh, xyxy, conf, cls in detections:
if isinstance(conf, tuple):
conf = conf[0]
img_lab = f'{conf:.2f}'
# Add bbox to image
c = int(cls) # integer class
label = True
if first and first[c]:
first[c] = 0
else:
label = False
name = f'{self.names[c]} ' if label else ''
label = f'{name}{img_lab}'
plot_one_box(xyxy,
img_out,
label=label,
color=colors(c, True),
line_thickness=6)
return img_out
def run_detections(self, opt):
image_files = FileIterator(opt.source)
for k, img_file in enumerate(image_files):
p = Path(img_file) # to Path
if opt.save_dir is None:
print('save_dir is set to None!')
sys.exit()
t1 = time_synchronized()
detections, img0 = self.get_detections(img_file) #, opt)
t2 = time_synchronized()
if k % 10 == 0: print(k)
if opt.skip_empty and len(detections) == 0:
continue
# Print results to screen(?)
s = '%gx%g ' % img0.shape[:2] # print string
cls_idx = np.array([det[-1].cpu().numpy() for det in detections])
for c in np.unique(cls_idx):
n = (cls_idx == c).sum() # detections per class
s += f"{n} {self.names[int(c)]}{'s' * (n > 1)}, " # add to string
save_path = str(opt.save_dir / p.name)
txt_path = str(opt.save_dir / 'labels' / p.stem)
first = self.names.copy() if self.names else None
# Loop over detections
imc = img0.copy() if opt.save_crop else img0 # for opt.save_crop
for xywh, xyxy, conf, cls in detections:
# Write label to txt file
txt_lab, img_lab = self.format_labels(xywh, cls, conf, opt)
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * len(txt_lab)).rstrip() % txt_lab + '\n')
# Add bbox to image
if opt.save_img or opt.save_crop:
c = int(cls) # integer class
label = True
if first and first[c]:
first[c] = 0
else:
label = False
name = f'{self.names[c]} ' if label else ''
label = None if opt.hide_labels else f'{name}{img_lab}'
plot_one_box(xyxy,
img0,
label=label,
color=colors(c, True),
line_thickness=opt.line_thickness)
if opt.save_crop:
save_one_box(xyxy,
imc,
file=opt.save_dir / 'crops' /
self.names[c] / f'{p.stem}.jpg',
BGR=True)
print(f'{s}Done. ({t2 - t1:.3f}s)')
if opt.save_img and len(detections) > 0:
cv2.imwrite(save_path, img0)
if len(detections) == 0:
empty_file(txt_path + '.txt')
def detect(opt):
source, weights, view_img, imgsz = opt.source, opt.weights, opt.view_img, opt.img_size
save_dir = 'output'
file_name = Path(source).name
source = cv2.VideoCapture(source)
width = int(source.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(source.get(cv2.CAP_PROP_FRAME_HEIGHT))
video_writer = cv2.VideoWriter('video_detection.avi', cv2.VideoWriter_fourcc(*'MJPG'), 20, (width, height))
while True:
validation, frame = source.read()
if validation is not True:
break
# Load model
print(validation)
device = select_device(opt.device)
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 device.type != 'cpu':
model(torch.zeros(1, 3, imgsz, imgsz).to(device).type_as(next(model.parameters()))) # run once
img = frame
original_image = img.copy()
t0 = time.time()
# img = cv2.resize(img, (416, 416))
img = letterbox(img)[0]
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img)
img = torch.from_numpy(img).to(device)
img = img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
img = img.unsqueeze(0)
# Inference
pred = model(img, augment=False)[0]
# Apply NMS
pred = non_max_suppression(pred, opt.conf_thres, opt.iou_thres, classes=opt.classes, agnostic=opt.agnostic_nms)
save_path = os.path.join(save_dir, file_name)
# Process detections
for det in pred: # detections per image
if len(det):
# Rescale boxes from img size to original_image size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4], original_image.shape).round()
# Write results
for *xyxy, conf, cls in reversed(det):
c = int(cls) # integer class
label = (names[c] if opt.hide_conf else f'{names[c]} {conf:.2f}')
plot_one_box(xyxy, original_image, label=label, color=colors(c, True), line_thickness=2)
# if view_img:
cv2.imshow("result", original_image)
cv2.waitKey(1) # 1 millisecond
# Save results (image with detections)
# cv2.imwrite(save_path, original_image)
# video_writer.write(original_image)
# video_writer.release()
print(f'Done. ({time.time() - t0:.3f}s)')
def run(
weights=ROOT / 'yolov5s.pt', # model.pt path(s) 训练的权重
source=ROOT /
'data/images', # file/dir/URL/glob, 0 for webcam 测试数据，图片/视频路径，'0'摄像头，rtsp视频流
imgsz=640, # inference size (pixels) 网络输入图片大小
conf_thres=0.25, # confidence threshold 置信度阈值
iou_thres=0.45, # NMS IOU threshold nms的iou阈值
max_det=1000, # maximum detections per image 分类数
device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu 设备
view_img=True, # show results 是否展示预测之后的图片/视频
save_txt=False, # save results to *.txt 是否将预测的框坐标保持txt格式，默认false
# save_conf=False, # save confidences in --save-txt labels 置信度保存
save_crop=False, # save cropped prediction boxes
nosave=False, # do not save images/videos 不保存
classes=None, # filter by class: --class 0, or --class 0 2 3 设置只保留某一部分类别
agnostic_nms=False, # class-agnostic NMS 进行nms是否也去除不同类别之间的框
augment=False, # augmented inference 图像增强
visualize=False, # visualize features 可视化
# update=False, # update all models 若ture，则对所有模型进行strip_optimizer操作，去除pt文件中的优化器等信息，默认false
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
):
source = str(source)
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 加载float32模型，确保图片分辨率能整除32
stride = int(model.stride.max()) # model stride
names = model.module.names if hasattr(
model, 'module') else model.names # get class names
#设置Float16
if half:
model.half() # to FP16
# 设置2次分类
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
# Dataloader
# 通过不同的输入源来设置不同的数据加载方式
# 摄像头
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
# 图片或视频
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':
# 进行一次前向推理，测试程序是否正常
model(
torch.zeros(1, 3, *imgsz).to(device).type_as(
next(model.parameters()))) # run once
dt, seen = [0.0, 0.0, 0.0], 0
'''
path 图片/视频路径
img 进行resize+pad之后的图片，如(3,640,512) 格式(c,h,w)
img0s 原size图片，如(1080,810,3)
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:
t1 = time_sync()
if onnx:
img = img.astype('float32')
else:
img = torch.from_numpy(img).to(device)
# 图片也设置为Float16或者32
img = img.half() if half else img.float() # uint8 to fp16/32
img = img / 255.0 # 0 - 255 to 0.0 - 1.0
# 没有batch_size时，在最前面添加一个轴
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
'''
前向传播，返回pred的shape是(1,num_boxes,5+num_class)
h,w为传入网络图片的高和宽，注意dataset在检测时使用了矩形推理，所以h不一定等于w
num_boxes = (h/32*w/32+h/16*w/16+h/8*w/8)*3
例如：图片大小720，1280 -> 15120个boxes = (20*12 + 40*24 + 80*48 = 5040)*3
pred[...,0:4]为预测框坐标；预测框坐标为xywh
pred[...,4]为objectness置信度
pred[...,5:-1]为分类结果
'''
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: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_nmsL进行nms是否也去除不同类别之间的框
经过nms后预测框格式，xywh->xyxy(左上角右上角)
pred是一个列表list[torch.tensor],长度为nms后目标框个数
每一个torch.tensor的shape为(num_boxes,6),内容为box(4个值)+cunf+cls
'''
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)
# 添加二级分类，默认false
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process predictions
# 对每一张图片处理
for i, det in enumerate(pred): # per image
seen += 1
# 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], f'{i}: ', im0s[i].copy(
), dataset.count
else:
p, s, im0, frame = path, '', im0s.copy(), getattr(
dataset, 'frame', 0)
p = Path(p) # to Path
# 设置保存图片或视频的路径
# p是原图片路径
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（图片宽高），s如'640*512'
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,
example=str(names))
if len(det):
# Rescale boxes from img_size to im0 size
# 调整预测框坐标，基于resize+pad的图片坐标->基于原size图片坐标
# 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' * (n > 1)}, " # 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格式转为xywh格式，并除上我w，h作归一化，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 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'{pred[0][0][0].tolist()} {pred[0][0][1].tolist()} {s}Done. ({t3 - t2:.3f}s)')
# Stream results
im0 = annotator.result()
# xxx = (pred[0][0][0].tolist()+pred[0][0][2].tolist())/2
# yyy = (pred[0][0][1].tolist()+pred[0][0][3].tolist())/2
if view_img:
# + / 2 +
cv2.imshow(str(p), im0)
cv2.moveWindow(str(p), 0, 0)
# pyautogui.moveTo(xxx, yyy)
cv2.waitKey(1000) # 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)
def run(
weights=ROOT / 'yolov5s.pt', # model.pt path(s)
source=ROOT / '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=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
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 = str(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 = torch.jit.load(w) if 'torchscript' in w else attempt_load(
weights, map_location=device, fuse=False)
stride = int(model.stride.max()) # model stride
names = model.module.names if hasattr(
model, 'module') else model.names # get class names
"""
for _, param in enumerate(model.named_parameters()):
print("====>", param[0], param[1].shape)
torch.save(model.state_dict(), 'new_params.pt')
for k, v in model.state_dict().items():
print(k, v.shape)
exit()
"""
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:
if dnn:
# check_requirements(('opencv-python>=4.5.4',))
net = cv2.dnn.readNetFromONNX(w)
else:
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
# 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]
anchor_grid = model.model[-1].anchors * model.model[-1].stride[
..., None, None]
delattr(model.model[-1],
'anchor_grid') # model.model[-1] is detect layer
model.model[-1].register_buffer("anchor_grid", anchor_grid)
model.to(device).eval()
wts_file = "generated.wts"
with open(wts_file, 'w') as f:
f.write('{}\n'.format(len(model.state_dict().keys())))
for k, v in model.state_dict().items():
if len(v.shape) == 0:
continue
print(k, v.shape)
vr = v.reshape(-1).cpu().numpy()
f.write('{} {} {} {}'.format(
k, len(vr), v.shape[0],
v.shape[1] if len(v.shape) > 1 else 0))
for vv in vr:
f.write(' ')
f.write(struct.pack('>f', float(vv)).hex())
f.write('\n')
exit()
elif onnx:
if dnn:
net.setInput(img)
pred = torch.tensor(net.forward())
else:
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,
example=str(names))
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(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], '%g: ' % 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
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,
im0s,
file=save_dir / 'crops' / names[c] /
f'{p.stem}.jpg',
BGR=True)
# 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' 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)
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 run(
weights=ROOT / 'yolov5s.pt', # model.pt path(s)
source=ROOT / 'data/images', # file/dir/URL/glob, 0 for webcam
data=ROOT / 'data/coco128.yaml', # dataset.yaml path
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,
data=data,
fp16=half)
stride, names, pt = model.stride, model.names, model.pt
imgsz = check_img_size(imgsz, s=stride) # check image size
# 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 if pt else bs, 3, *imgsz)) # 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 model.fp16 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(f'{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)
# 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 = str(Path(save_path).with_suffix(
'.mp4')) # force *.mp4 suffix on results videos
vid_writer[i] = cv2.VideoWriter(
save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps,
(w, h))
vid_writer[i].write(im0)
# Print time (inference-only)
LOGGER.info(f'{s}Done. ({t3 - t2:.3f}s)')
# 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(
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
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://'))
realsense = "realsense" in source
# 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
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 image 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='resnet50', n=2) # initialize
modelc.load_state_dict(
torch.load('resnet50.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)
elif realsense:
view_img = check_imshow()
cudnn.benchmark = True # set True to speed up constant image size inference
dataset = LoadRealSense2(width=640, height=480, fps=15, img_size=imgsz)
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, depth, distance, depth_scale, 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,
agnostic_nms,
max_det=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
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
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):
# Splitting xyxy* (measurement)
xmin = int(xyxy[0])
ymin = int(xyxy[1])
xmax = int(xyxy[2])
ymax = int(xyxy[3])
# Calculating depth of centroid of the object (in Pixel)
xc = int(round(((xmax + xmin) / 2), 0))
yc = int(round(((ymax + ymin) / 2), 0))
object_depth = depth[xc, yc]
object_distance = distance[xc, yc]
print("object depth:")
print(object_depth)
print("object distance:")
print(object_distance)
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}')
plot_one_box(xyxy,
im0,
label=label,
color=colors(c, True),
line_thickness=line_thickness)
if 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)')
# 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)
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}")
if update:
strip_optimizer(weights) # update model (to fix SourceChangeWarning)
print(f'Done. ({time.time() - t0:.3f}s)')
def apply(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
# 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
(save_dir / 'data' if save_txt else save_dir).mkdir(parents=True, exist_ok=True)
with (save_dir / f"params_{Path(opt.source).name}.json").open("w") as f:
f.write(json.dumps(opt.__dict__, indent=4))
# 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
# Set Dataloader
vid_path, vid_writer = None, None
dataset = LoadRiceImages(source, img_size=imgsz, stride=stride, dshape=opt.dshape, ishape=opt.ishape)
# 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()
my = None
for path, imgs, imgs0, _, big_img in dataset:
path = Path(path)
ori_img = cv2.imread(str(path))
save_path = str(save_dir / path.name)
txt_path = str(save_dir / "labels" / f"{path.stem}.csv")
data_path = str(save_dir / "data" / f"{path.stem}.csv")
coords = []
img_type = str(path.name)[0].lower()
for r in range(imgs.shape[0]):
for c in range(imgs.shape[1]):
conf_thres = opt.i_conf_thres if img_type == "i" else opt.d_conf_thres
img = imgs[r, c]
im0s = imgs0[r, c]
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, 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, frame = path, '', im0s.copy(), 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 cl in det[:, -1].unique():
n = (det[:, -1] == cl).sum() # detections per class
s += f"{n} {names[int(cl)]}{'s' * (n > 1)}, " # add to string
# Write results
for *xyxy, conf, cl in reversed(det):
print(xyxy)
# sys.exit(0)
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
x, y = xywh[:2]
x, y = x * im0.shape[1], y * im0.shape[0]
x += c * im0.shape[1]
y += r * im0.shape[0]
cl = cl.cpu()
# Only append if the predicted class matches the img_type
if (cl == 0 and img_type == "i") or (cl == 1 and img_type == "d"):
coords.append(np.array((conf.cpu().item() * 100, x, y, cl)))
if save_img or view_img: # Add bbox to image
c = int(cl) # 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)
# Print time (inference + NMS)
print(f'{s}Done. ({t2 - t1:.3f}s)')
# Save results (image with detections)
if save_img:
if dataset.mode == 'image':
cv2.imwrite('out_old.jpg', im0)
sys.exit(0)
v_grid_starts, h_grid_starts = [], []
if opt.grid:
x = imgs[0, 0].shape[2]
while x < ori_img.shape[1]:
v_grid_starts.append(x)
x += imgs[0, 0].shape[2]
y = imgs[0, 0].shape[1]
while y < ori_img.shape[0]:
h_grid_starts.append(y)
y += imgs[0, 0].shape[1]
v_grid_starts, h_grid_starts = np.array(v_grid_starts, dtype=float), np.array(h_grid_starts, dtype=float)
# imgs[0, 0].shape is (c, h, w)
scale_x = ori_img.shape[1] / (imgs.shape[1] * imgs[0, 0].shape[2])
scale_y = ori_img.shape[0] / (imgs.shape[0] * imgs[0, 0].shape[1])
coords = np.array(coords)
coords[:, 1] *= scale_x
coords[:, 2] *= scale_y
coords = np.around(coords).astype(int)
close_tol = opt.i_close if img_type == "i" else opt.d_close
v_grid_starts *= scale_x
h_grid_starts *= scale_y
v_grid_starts, h_grid_starts = np.around(v_grid_starts).astype(int), np.around(h_grid_starts).astype(int)
axis_expand = opt.i_axis_expand if img_type == "i" else opt.d_axis_expand
coords = filter_too_close(coords, tolerance=close_tol, h_axis=h_grid_starts, v_axis=v_grid_starts, axis_expand=axis_expand)
coords = filter_border(coords, ori_img.shape, tolerance=opt.border)
gt_path = path.parent / f"{path.stem}.csv"
if save_txt:
with open(txt_path, "w") as f:
np.savetxt(f, coords[:, 1:3], fmt="%d", delimiter=",")
with open(data_path, "w") as f:
np.savetxt(f, coords[:, 0:3], fmt="%d", delimiter=",")
if save_img:
if "border" in vars(opt) and opt.border > 0:
ori_img = draw_border(ori_img, opt.border)
if opt.grid:
ori_img = draw_grid(ori_img, v_grid_starts, h_grid_starts)
if opt.with_gt:
gts = np.loadtxt(gt_path, dtype=int, delimiter=",", ndmin=2)
for x, y in gts:
ori_img = cv2.circle(ori_img, (x, y), 9, (255, 255, 255), 2)
for conf, x, y, cl in coords:
if cl == 0:
circle_color = (255, 0, 0)
elif cl == 1:
circle_color = (0, 0, 255)
if not opt.hide_conf:
# print(conf)
ori_img = cv2.putText(ori_img, f"{conf}%", (x, y - 3), 0, 1, (255, 255, 0), 2)
ori_img = cv2.circle(ori_img, (x, y), 4, circle_color, -1)
cv2.imwrite(save_path, ori_img)
if save_txt or save_img:
s = f"\n{len(list(save_dir.glob('labels/*')))} 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_image(
self,
img_cv,
conf_thres=0.25, # confidence threshold
iou_thres=0.45, # NMS IOU threshold
max_det=1000, # maximum detections per image
classes=None,
agnostic_nms=False, # class-agnostic NMS
line_thickness=3, # bounding box thickness (pixels)
hide_labels=False, # hide labels
hide_conf=False, # hide confidences
):
device = select_device('')
### Preprocess based on what inside LoadImages
# Padded resize
img = letterbox(img_cv, 640, stride=32)[0]
# Convert
img = img.transpose((2, 0, 1))[::-1] # HWC to CHW, BGR to RGB
img = np.ascontiguousarray(img)
### Preprocess finish
img = torch.from_numpy(img).to(device)
img = img.float()
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if len(img.shape) == 3:
img = img[None] # expand for batch dim
# Inference
pred = self.model(img, augment=False, visualize=False)[0]
# NMS
pred = non_max_suppression(pred,
conf_thres,
iou_thres,
classes,
agnostic_nms,
max_det=max_det)
# Process predictions
bbox_pred = []
s, im0 = '', img_cv.copy()
for i, det in enumerate(pred): # detections per image
s += '%gx%g ' % img.shape[2:] # print string
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Visualize results
for *xyxy, conf, cls in reversed(det):
c = int(cls) # integer class
label = None if hide_labels else (
self.names[c]
if hide_conf else f'{self.names[c]} {conf:.2f}')
plot_one_box(xyxy,
im0,
label=label,
color=colors(c, True),
line_thickness=line_thickness)
bbox_pred.append(
det.cpu().numpy()[:, :-1]
) # bbox pred after rescaled back to the original image size
return bbox_pred[0], im0
0]] # normalization gain whwh
for i, det in enumerate(pred):
if len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
out_img.shape).round()
# Write results
for *xyxy, conf, cls in reversed(det):
if opt.show: # Add bbox to image
c = int(cls) # integer class
label = f'{names[c]} {conf:.2f}'
plot_one_box(xyxy,
out_img,
label=label,
color=colors(c, True),
line_thickness=opt.line_thickness)
t1 = time.time()
t2 = time.time()
times_infer.append(t1 - t0)
times_pipe.append(t2 - t0)
times_infer = times_infer[-20:]
times_pipe = times_pipe[-20:]
ms = sum(times_infer) / len(times_infer) * 1000
fps_infer = 1000 / (ms + 0.00001)
fps_pipe = 1000 / (sum(times_pipe) / len(times_pipe) * 1000)
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://'))
sheet = pd.DataFrame()
# 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, pt, onnx = False, w.endswith('.pt'), w.endswith('.onnx') # inference type
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)
imgsz = check_img_size(imgsz, s=stride) # check image size
# 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)
bs = len(dataset) # batch_size
else:
dataset = LoadImages(source, img_size=imgsz, stride=stride)
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, imgsz).to(device).type_as(next(model.parameters()))) # run once
t0 = time.time()
for path, img, im0s, vid_cap in dataset:
# print(path)
# sys.exit(0)
if pt:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
elif onnx:
img = img.astype('float32')
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if len(img.shape) == 3:
img = img[None] # expand for batch dim
# Inference
t1 = time_sync()
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}))
# NMS
pred = non_max_suppression(pred, conf_thres, iou_thres, classes, agnostic_nms, max_det=max_det)
t2 = time_sync()
# Second-stage classifier (optional)
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
# Process predictions
# print(len(pred))
for i, det in enumerate(pred): # detections per image
# print(det.shape)
# sys.exit(0)
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
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')
lst = torch.tensor(xyxy).tolist()
# print(i)
dic = dict(name=names[int(cls)], image_id=str(path.split('\\')[-1].split('.')[0]), confidence=float(conf), xmin=int(lst[0]), ymin=int(lst[1]), xmax=int(lst[2]), ymax=int(lst[3]))
sheet = sheet.append(dic, ignore_index=True)
# print(dic)
# sys.exit(0)
# print(dic)
# sys.exit(0)
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}')
plot_one_box(xyxy, im0, label=label, color=colors(c, True), line_thickness=line_thickness)
if save_crop:
save_one_box(xyxy, imc, file=save_dir / 'crops' / names[c] / f'{p.stem}.jpg', BGR=True)
# print(sheet)
# sys.exit(0)
# 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' 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)
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}")
if update:
strip_optimizer(weights) # update model (to fix SourceChangeWarning)
cols=['name', 'image_id', 'confidence', 'xmin', 'ymin', 'xmax', 'ymax']
sheet = sheet.loc[:, cols]
# sheet = sheet.set_index('name')
sheet.to_csv('result.csv', index=False)
print(f'Done. ({time.time() - t0:.3f}s)')
def run(
weights=ROOT / 'yolov5s.pt', # model.pt path(s) 训练的权重
imgsz=[640, 640], # inference size (pixels) 网络输入图片大小
conf_thres=0.25, # confidence threshold 置信度阈值
iou_thres=0.45, # NMS IOU threshold nms的iou阈值
max_det=1000, # maximum detections per image 分类数
device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu 设备
view_img=True, # show results 是否展示预测之后的图片/视频
classes=None, # filter by class: --class 0, or --class 0 2 3 设置只保留某一部分类别
agnostic_nms=False, # class-agnostic NMS 进行nms是否也去除不同类别之间的框
augment=False, # augmented inference 图像增强
visualize=False, # visualize features 可视化
line_thickness=3, # bounding box thickness (pixels)
hide_labels=False, # hide labels
hide_conf=False, # hide confidences
half=False, # use FP16 half-precision inference
):
# 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 加载float32模型，确保图片分辨率能整除32
stride = int(model.stride.max()) # model stride
names = model.module.names if hasattr(
model, 'module') else model.names # get class names
#设置Float16
if half:
model.half() # to FP16
# 设置2次分类
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()
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
# Dataloader
# 图片或视频
tmp = False
tmp2 = False
mon = {'top': 0, 'left': 0, 'width': 960, 'height': 960}
while True:
im = np.array(mss().grab(mon))
screen = cv2.cvtColor(im, cv2.COLOR_BGRA2BGR)
dataset = LoadImages(screen, img_size=imgsz, stride=stride, auto=pt)
dt, seen = [0.0, 0.0, 0.0], 0
'''
path 图片/视频路径
img 进行resize+pad之后的图片，如(3,640,512) 格式(c,h,w)
img0s 原size图片，如(1080,810,3)
cap 当读取图片时为None,读取视频时为视频源
'''
for img, im0s, vid_cap in dataset:
t1 = time_sync()
if onnx:
img = img.astype('float32')
else:
img = torch.from_numpy(img).to(device)
# print(img)
# 图片也设置为Float16或者32
img = img.half() if half else img.float() # uint8 to fp16/32
img = img / 255.0 # 0 - 255 to 0.0 - 1.0
# 没有batch_size时，在最前面添加一个轴
if len(img.shape) == 3:
img = img[None] # expand for batch dim
t2 = time_sync()
dt[0] += t2 - t1
# Inference
if pt:
'''
前向传播，返回pred的shape是(1,num_boxes,5+num_class)
h,w为传入网络图片的高和宽，注意dataset在检测时使用了矩形推理，所以h不一定等于w
num_boxes = (h/32*w/32+h/16*w/16+h/8*w/8)*3
例如：图片大小720，1280 -> 15120个boxes = (20*12 + 40*24 + 80*48 = 5040)*3
pred[...,0:4]为预测框坐标；预测框坐标为xywh
pred[...,4]为objectness置信度
pred[...,5:-1]为分类结果
'''
pred = model(img, augment=augment, visualize=visualize)[0]
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: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_nmsL进行nms是否也去除不同类别之间的框
经过nms后预测框格式，xywh->xyxy(左上角右上角)
pred是一个列表list[torch.tensor],长度为nms后目标框个数
每一个torch.tensor的shape为(num_boxes,6),内容为box(4个值)+cunf+cls
'''
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)
# 添加二级分类，默认false
# if classify:
# pred = apply_classifier(pred, modelc, img, im0s)
# Process predictions
# 对每一张图片处理
for i, det in enumerate(pred): # per image
seen += 1
s, im0 = '', im0s.copy()
# 设置打印信息（图片宽高），s如'640*512'
s += '%gx%g ' % img.shape[2:] # print string
annotator = Annotator(im0,
line_width=line_thickness,
example=str(names))
if len(det):
# Rescale boxes from img_size to im0 size
# 调整预测框坐标，基于resize+pad的图片坐标->基于原size图片坐标
# 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' * (n > 1)}, " # add to string
# Write results
# 保存预测结果
for *xyxy, conf, cls in reversed(det):
if 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))
# Stream results
im0 = annotator.result()
cv2.imshow('a crop of the screen', im0)
cv2.moveWindow('a crop of the screen', 960, 0)
if cv2.waitKey(1) & 0xff == ord('q'):
tmp = True
break
if tmp:
tmp2 = True
break
if tmp2:
break
gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
imc = img0.copy() # for save_crop
annotator = Annotator(im0, line_width=line_thickness)
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):
# 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}')
label= None
annotator.box_label(xyxy, label, color=colors(c, True))
# Stream results
im0 = annotator.result()
if True:
cv2.imwrite("pred.jpg", im0)
cv2.waitKey(0) # 1 millisecond
def run(
weights='yolov5s.pt', # model.pt path(s)
source='./test_1', # 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='', # = 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 labelss
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
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
):
desire_param = []
coor = []
all_info = []
save_img = not nosave and not source.endswith(
'.txt') # save inference images
# 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
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 image 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='resnet50', n=2) # initialize
modelc.load_state_dict(
torch.load('resnet50.pt',
map_location=device)['model']).to(device).eval()
# Set Dataloader
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=augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
conf_thres,
iou_thres,
classes,
agnostic_nms,
max_det=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
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
#print()
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
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
pr = ' '
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
pr += f"{n} {names[int(c)]}{'s' * (n > 1)},"
desire_param.append({"image_id": p.name, "prediction": pr})
# for img_name in enumerate(p.name):
# if img_name not in desire_param:
# add_image = (img_name, "predictioni 0")
# desire_param.append(add_image)
# print(desire_param)
# 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}')
plot_one_box(xyxy,
im0,
label=label,
color=colors(c, True),
line_thickness=line_thickness)
if 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)')
result = [(f'{s}Done. ({t2 - t1:.3f}s)')]
print(result)
# 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:
None
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}")
if update:
strip_optimizer(weights) # update model (to fix SourceChangeWarning)
print(f'Done. ({time.time() - t0:.3f}s)')
# lst = []
# for child in desire_param:
# info = ["img_name", "prediction"]
# lst1 = {k: v for k, v in zip(info, child)}
# lst.append(lst1)
with open('result.json', 'w') as f:
json.dump(desire_param, f)
def display(self,
pprint=False,
show=False,
save=False,
crop=False,
render=False,
labels=True,
save_dir=Path('')):
crops = []
for i, (im, pred) in enumerate(zip(self.imgs, self.pred)):
s = f'image {i + 1}/{len(self.pred)}: {im.shape[0]}x{im.shape[1]} ' # string
if pred.shape[0]:
for c in pred[:, -1].unique():
n = (pred[:, -1] == c).sum() # detections per class
s += f"{n} {self.names[int(c)]}{'s' * (n > 1)}, " # add to string
if show or save or render or crop:
annotator = Annotator(im, example=str(self.names))
for *box, conf, cls in reversed(
pred): # xyxy, confidence, class
label = f'{self.names[int(cls)]} {conf:.2f}'
if crop:
file = save_dir / 'crops' / self.names[int(
cls)] / self.files[i] if save else None
crops.append({
'box':
box,
'conf':
conf,
'cls':
cls,
'label':
label,
'im':
save_one_box(box, im, file=file, save=save)
})
else: # all others
annotator.box_label(box,
label if labels else '',
color=colors(cls))
im = annotator.im
else:
s += '(no detections)'
im = Image.fromarray(im.astype(np.uint8)) if isinstance(
im, np.ndarray) else im # from np
if pprint:
print(s.rstrip(', '))
if show:
im.show(self.files[i]) # show
if save:
f = self.files[i]
im.save(save_dir / f) # save
if i == self.n - 1:
LOGGER.info(
f"Saved {self.n} image{'s' * (self.n > 1)} to {colorstr('bold', save_dir)}"
)
if render:
self.imgs[i] = np.asarray(im)
if crop:
if save:
LOGGER.info(f'Saved results to {save_dir}\n')
return crops
def detect(path_request):
parser = argparse.ArgumentParser()
parser.add_argument('--weights', nargs='+', type=str, default='last_ufpr.pt', help='model.pt path(s)')
parser.add_argument('--source', type=str, default='data/images', help='source') # file/folder, 0 for webcam
parser.add_argument('--img-size', type=int, default=640, help='inference size (pixels)')
parser.add_argument('--conf-thres', type=float, default=0.25, help='object confidence threshold')
parser.add_argument('--iou-thres', type=float, default=0.45, help='IOU threshold for NMS')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--view-img', action='store_true', help='display results')
parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
parser.add_argument('--save-crop', action='store_true', help='save cropped prediction boxes')
parser.add_argument('--nosave', action='store_true', help='do not save images/videos')
parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --class 0, or --class 0 2 3')
parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS')
parser.add_argument('--augment', action='store_true', help='augmented inference')
parser.add_argument('--update', action='store_true', help='update all models')
parser.add_argument('--project', default='runs/detect', help='save results to project/name')
parser.add_argument('--name', default='exp', help='save results to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--line-thickness', default=3, type=int, help='bounding box thickness (pixels)')
parser.add_argument('--hide-labels', default=False, action='store_true', help='hide labels')
parser.add_argument('--hide-conf', default=False, action='store_true', help='hide confidences')
opt = parser.parse_args()
#print(opt)
check_requirements(exclude=('tensorboard', 'pycocotools', 'thop'))
source, weights, view_img, save_txt, imgsz = path_request, opt.weights, opt.view_img, opt.save_txt, opt.img_size
save_img = not opt.nosave and not source.endswith('.txt') # save inference images
dataset = LoadImages(source, img_size=imgsz, stride=stride)
webcam = False
# 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
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)
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)
#print("Predicted",pred)
# Process detections
dict_prediction = {}
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
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()
ans = ""
class_cat = []
class_conf = []
for i,c in enumerate(det[:, -1]):
if names[int(c)] == "car" or names[int(c)] == "plate" or names[int(c)] == "motorcycle":
ans+=f"{names[int(c)]}, "
for i,c in enumerate(det[:, -1]):
class_cat.append(int(c))
for i,c in enumerate(det[:, -2]):
class_conf.append(float(c))
print("all",class_cat,class_conf)
# Print results
for c in det[:, -1].unique():
#print(c)
n = (det[:, -1] == c).sum() # detections per class
s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
#print("results:",s)
# 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)
# dict_prediction["class_cat"] = "".join((str(value) for value in class_cat))
print("ans",ans)
dict_prediction["class_cat"] = ans
dict_prediction["class_conf"] = "".join((str(value) for value in class_conf))
json_data = json.dumps(dict_prediction)
return json_data
# if __name__ == '__main__':
# parser = argparse.ArgumentParser()
# parser.add_argument('--weights', nargs='+', type=str, default='yolov5s.pt', help='model.pt path(s)')
# parser.add_argument('--source', type=str, default='data/images', help='source') # file/folder, 0 for webcam
# parser.add_argument('--img-size', type=int, default=640, help='inference size (pixels)')
# parser.add_argument('--conf-thres', type=float, default=0.25, help='object confidence threshold')
# parser.add_argument('--iou-thres', type=float, default=0.45, help='IOU threshold for NMS')
# parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
# parser.add_argument('--view-img', action='store_true', help='display results')
# parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
# parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
# parser.add_argument('--save-crop', action='store_true', help='save cropped prediction boxes')
# parser.add_argument('--nosave', action='store_true', help='do not save images/videos')
# parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --class 0, or --class 0 2 3')
# parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS')
# parser.add_argument('--augment', action='store_true', help='augmented inference')
# parser.add_argument('--update', action='store_true', help='update all models')
# parser.add_argument('--project', default='runs/detect', help='save results to project/name')
# parser.add_argument('--name', default='exp', help='save results to project/name')
# parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
# parser.add_argument('--line-thickness', default=3, type=int, help='bounding box thickness (pixels)')
# parser.add_argument('--hide-labels', default=False, action='store_true', help='hide labels')
# parser.add_argument('--hide-conf', default=False, action='store_true', help='hide confidences')
# opt = parser.parse_args()
# print(opt)
# check_requirements(exclude=('tensorboard', 'pycocotools', 'thop'))
# with torch.no_grad():
# if opt.update: # update all models (to fix SourceChangeWarning)
# for opt.weights in ['yolov5s.pt', 'yolov5m.pt', 'yolov5l.pt', 'yolov5x.pt']:
# detect(opt=opt)
# strip_optimizer(opt.weights)
# else:
# detect(opt=opt)
#model.load_state_dict(model['state_dict'])
### now you can evaluate it
#model.eval()
#model.eval()
def detect(
model="mobilenet_thin", # A model option for being cool
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
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
):
w, h = 432, 368
e = TfPoseEstimator(get_graph_path(model), target_size=(w, h))
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 = Path(project)
#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
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 image 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
breakCond = False
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)
# Openpose getting keypoints and individual crops
print("\n")
myImg = im0s.copy()
keypoints, humans = getKeyPoints(myImg, e, w, h)
crops = [
getCrop(point[0], myImg, 10, device, point[1] / 2)
for point in keypoints
]
# Inference
t1 = time_synchronized()
pred = model(img, augment=augment)[0]
# Apply NMS
pred = non_max_suppression(pred,
conf_thres,
iou_thres,
classes,
agnostic_nms,
max_det=max_det)
t2 = time_synchronized()
# Need to adjust bboxes to full image
if len(pred) > 0:
breakCond = True
# 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
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 save_crop else im0 # for 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
# Check if any overlap between keypoint and det (handheld weapon)
for detection in det:
for crop in crops:
if bbox_iou(detection, crop) > 0:
cv2.putText(im0, "Spider-Sense Tingling!",
(30, 90), cv2.FONT_HERSHEY_SIMPLEX, 3,
(255, 0, 0), 5)
break
# 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}')
plot_one_box(xyxy,
im0,
label=label,
color=colors(c, True),
line_thickness=line_thickness)
if save_crop:
save_one_box(xyxy,
imc,
file=save_dir / 'crops' / names[c] /
f'{p.stem}.jpg',
BGR=True)
# write keypoint boxes
for *xyxy, conf, cls in reversed(crops):
plot_one_box(xyxy,
imc,
label="keyP",
color=colors(c, True),
line_thickness=line_thickness)
# Stream results
if view_img:
cv2.imshow(str(p), im0)
cv2.waitKey(1) # 1 millisecond
# Save results (image with detections)
im0 = TfPoseEstimator.draw_humans(im0, humans, imgcopy=False)
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)
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}")
if update:
strip_optimizer(weights) # update model (to fix SourceChangeWarning)
print(f'Done. ({time.time() - t0:.3f}s)')
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 ''