def detect( self, image_data ):
input_image = image_data.asarray().astype( 'uint8' )
from mmdet.apis import inference_detector
gpu_string = 'cuda:' + str( self._gpu_index )
detections = inference_detector( self._model, input_image, self._cfg, device=gpu_string )
class_names = [ 'fish' ] * 10000
if isinstance( detections, tuple ):
bbox_result, segm_result = detections
else:
bbox_result, segm_result = detections, None
if np.size( bbox_result ) > 0:
bboxes = np.vstack( bbox_result )
else:
bboxes = []
sys.stdout.write( "Detected " + str( len( bbox_result ) ) + " objects" )
sys.stdout.flush()
# convert segmentation masks
masks = []
if segm_result is not None:
segms = mmcv.concat_list( segm_result )
inds = np.where( bboxes[:, -1] > score_thr )[0]
for i in inds:
masks.append( maskUtils.decode( segms[i] ).astype( np.bool ) )
# collect labels
labels = [
np.full( bbox.shape[0], i, dtype=np.int32 )
for i, bbox in enumerate( bbox_result )
]
if np.size( labels ) > 0:
labels = np.concatenate( labels )
else:
labels = []
# convert to kwiver format, apply threshold
output = []
for entry in []:
output.append( DetectedObject( BoundingBox( 1,1,2,2 ) ) )
if np.size( labels ) > 0:
mmcv.imshow_det_bboxes(
input_image,
bboxes,
labels,
class_names=class_names,
score_thr=-100.0,
show=True)
return DetectedObjectSet( output )
def result2json(checkpoint_file):
config_file = 'code/cascade_skypool_dataenhancement.py' # 修改成自己的配置文件
# checkpoint_file = 'work_dirs/cascade_custom_data_enhancement/epoch_56.pth' # 修改成自己的训练权重
test_path = 'data/guangdong1_round1_testA_20190818/testA' # 官方测试集图片路径
json_name = "result_" + "" + time.strftime("%Y%m%d%H%M%S",
time.localtime()) + ".json"
model = init_detector(config_file, checkpoint_file, device='cuda:0')
img_list = []
for img_name in os.listdir(test_path):
if img_name.endswith('.jpg'):
img_list.append(img_name)
result = []
for i, img_name in enumerate(img_list, 1):
full_img = os.path.join(test_path, img_name)
predict = inference_detector(model, full_img)
for i, bboxes in enumerate(predict, 1):
if len(bboxes) > 0:
# defect_label = lable[i]
# print(defect_label)
defect_label = i
print(i)
image_name = img_name
for bbox in bboxes:
x1, y1, x2, y2, score = bbox.tolist()
x1, y1, x2, y2 = round(x1, 2), round(y1, 2), round(
x2, 2), round(y2, 2) # save 0.00
result.append({
'name': image_name,
'category': defect_label,
'bbox': [x1, y1, x2, y2],
'score': score
})
with open(json_name, 'w') as fp:
json.dump(result, fp, indent=4, separators=(',', ': '))
def inference_single(self, imagname, slide_size, chip_size):
img = mmcv.imread(imagname)
height, width, channel = img.shape
slide_h, slide_w = slide_size
hn, wn = chip_size
# TODO: check the corner case
# import pdb; pdb.set_trace()
total_detections = [
np.zeros((0, 9)) for _ in range(len(self.classnames))
]
for i in tqdm(range(int(width / slide_w + 1))):
for j in range(int(height / slide_h) + 1):
subimg = np.zeros((hn, wn, channel))
# print('i: ', i, 'j: ', j)
chip = img[j * slide_h:j * slide_h + hn,
i * slide_w:i * slide_w + wn, :3]
subimg[:chip.shape[0], :chip.shape[1], :] = chip
chip_detections = inference_detector(self.model, subimg)
# print('result: ', result)
for cls_id, name in enumerate(self.classnames):
chip_detections[cls_id][:, :8][:, ::2] = chip_detections[
cls_id][:, :8][:, ::2] + i * slide_w
chip_detections[cls_id][:, :8][:, 1::2] = chip_detections[
cls_id][:, :8][:, 1::2] + j * slide_h
# import pdb;pdb.set_trace()
try:
total_detections[cls_id] = np.concatenate(
(total_detections[cls_id],
chip_detections[cls_id]))
except:
import pdb
pdb.set_trace()
# nms
for i in range(len(self.classnames)):
keep = py_cpu_nms_poly_fast_np(total_detections[i], 0.1)
total_detections[i] = total_detections[i][keep]
return total_detections
def solo_infer(model, img, conf):
image_np = np.array(Image.open(img))
result, _ = inference_detector(model, img)
cur_result = result[0]
if cur_result is not None:
masks = cur_result[0].cpu().numpy().astype(np.uint8)
classes = cur_result[1].cpu().numpy()
scores = cur_result[2].cpu().numpy()
h, w = masks[0].shape
vis_inds = (scores > conf)
masks = masks[vis_inds]
classes = classes[vis_inds]
areas = [mask.sum() for mask in masks]
sorted_inds = np.argsort(areas)[::-1]
keep_inds = []
for i in sorted_inds:
if i != 0:
for j in range(i):
if np.sum((masks[i, :, :] > 0) * (masks[j, :, :] > 0)) / np.sum(masks[j, :, :] > 0) > 0.85:
break
keep_inds.append(i)
masks = masks[keep_inds]
classes = classes[keep_inds]
instance_map = np.zeros((h, w), dtype=np.uint8)
semantic_map = np.zeros((h, w), dtype=np.uint8)
if masks is not None:
for i, (mask, cls) in enumerate(zip(masks, classes)):
instance_map[mask > 0] = i + 1
semantic_map[mask > 0] = cls + 1
if cls in [0, 1, 7]:
color_mask = np.random.randint(0, 256, (1, 3), dtype=np.uint8)
mask_bool = mask.astype(np.bool)
image_np[mask_bool] = image_np[mask_bool] * 0.5 + color_mask * 0.5
final_mask = np.stack([instance_map, semantic_map], axis=-1)
return masks, classes, final_mask, image_np
def main():
parser = ArgumentParser()
# parser.add_argument('img', help='Image file')
parser.add_argument('config', help='Config file')
parser.add_argument('checkpoint', help='Checkpoint file')
parser.add_argument('--device',
default='cuda:0',
help='Device used for inference')
parser.add_argument('--score-thr',
type=float,
default=0.000000001,
help='bbox score threshold')
args = parser.parse_args()
# build the model from a config file and a checkpoint file
model = init_detector(args.config, args.checkpoint, device=args.device)
# test a single image
# import pdb;
# pdb.set_trace()
#
test_img_dir = "/home/dell/桌面/tile_round1_testA_20201231/tile_round1_testA_20201231/cam3/crop/"
imgs_dir = os.listdir(test_img_dir)
json_result = []
for img in imgs_dir:
img_name = img
img = test_img_dir + img
result = inference_detector(model, img)
# show the results
show_result_pyplot(model,
result,
img,
img_name,
json_result=json_result,
score_thr=args.score_thr)
with open("/home/dell/桌面/guangdong/result_cam3.json", "w") as f:
json.dump(json_result, f, cls=MyEncoder, indent=6)
print("加载入文件完成...")
def detect(self, img: np.ndarray) -> List[Detection]:
raw_result = inference_detector(self.model, img)
detections = []
if isinstance(raw_result, tuple):
for class_id in range(len(raw_result[0])):
box_result = raw_result[0][class_id]
mask_result = raw_result[1][class_id]
valid_indices = np.where(box_result[:, 4] > self.conf_thres)[0]
box_result = box_result[valid_indices]
mask_result = [mask_result[i] for i in valid_indices]
if mask_result != []:
# If mask result is empty, decode() will raise error
# mask_result = decode(mask_result).astype(np.bool)
detections.extend([
Detection(box_result[i][:4],
box_result[i][4],
class_id=class_id,
mask=mask_result[i])
for i in range(len(box_result))
])
detections.extend([
Detection(box_result[i][:4],
box_result[i][4],
class_id=class_id,
mask=None) for i in range(len(box_result))
])
else:
# Only boxes with class "person" are counted
for class_id in range(len(raw_result)):
class_result = raw_result[class_id]
class_result = class_result[np.where(
class_result[:, 4] > self.conf_thres)]
detections.extend([
Detection(line[:4], line[4], class_id=class_id)
for line in class_result
])
return detections
def predict(self, tasks, **kwargs):
assert len(tasks) == 1
task = tasks[0]
image_path = get_image_local_path(task['data'][self.value],
project_dir=self.project_path)
model_results = inference_detector(self.model, image_path)
results = []
all_scores = []
img_width, img_height = get_image_size(image_path)
for bboxes, label in zip(model_results, self.model.CLASSES):
output_label = self.label_map.get(label, label)
if output_label not in self.labels_in_config:
print(output_label + ' label not found in project config.')
continue
for bbox in bboxes:
bbox = list(bbox)
if not bbox:
continue
score = float(bbox[-1])
if score < self.score_thresh:
continue
x, y, xmax, ymax = bbox[:4]
results.append({
'from_name': self.from_name,
'to_name': self.to_name,
'type': 'rectanglelabels',
'value': {
'rectanglelabels': [output_label],
'x': x / img_width * 100,
'y': y / img_height * 100,
'width': (xmax - x) / img_width * 100,
'height': (ymax - y) / img_height * 100
},
'score': score
})
all_scores.append(score)
avg_score = sum(all_scores) / len(all_scores)
return [{'result': results, 'score': avg_score}]
def myfunc2(x):
idx, img = x
print('myfunc2: ', idx)
result = inference_detector(model, img)
print('len of share_submit_result:', len(share_submit_result))
for cls, item in enumerate(result):
if item is None:
continue
else:
for row in item:
# 获取锁
share_lock.acquire()
share_submit_result.append({
'name': img_name,
'category': cls + 1,
'bbox': row[:4].tolist(),
'score': str(row[4])
})
# 释放锁
share_lock.release()
def main():
parser = ArgumentParser()
parser.add_argument('img', help='Image file')
parser.add_argument('config', help='Config file')
parser.add_argument('checkpoint', help='Checkpoint file')
parser.add_argument('--device',
default='cuda:0',
help='Device used for inference')
parser.add_argument('--score-thr',
type=float,
default=0.3,
help='bbox score threshold')
args = parser.parse_args()
# build the model from a config file and a checkpoint file
model = init_detector(args.config, args.checkpoint, device=args.device)
# test a single image
result = inference_detector(model, args.img)
print("result is ", len(result))
for i in range(len(result)):
print("result ", i, " is ", len(result[i]))
print("get result!!!!!!!!!!!!!", result)
def main():
parser = ArgumentParser()
parser.add_argument('img', help='Image file')
parser.add_argument('config', help='Config file')
parser.add_argument('checkpoint', help='Checkpoint file')
parser.add_argument('--device',
default='cuda:0',
help='Device used for inference')
parser.add_argument('--score-thr',
type=float,
default=0.3,
help='bbox score threshold')
args = parser.parse_args()
# build the model from a config file and a checkpoint file
model = init_detector(args.config, args.checkpoint, device=args.device)
# test a single image
result = inference_detector(model, args.img)
# show the results
show_result_pyplot(args.img,
result, [model.CLASSES],
score_thr=args.score_thr)
def kitti():
cfg = mmcv.Config.fromfile('/nfs/project/libo_i/mmdetection/configs/KITTI/cascade_mask_rcnn_x101_64x4d_fpn_1x.py')
cfg.model.pretrained = None
# construct the model and load checkpoint
model = build_detector(cfg.model, test_cfg=cfg.test_cfg)
_ = load_checkpoint(model, '/nfs/project/libo_i/mmdetection/work_dirs/kitti_cascade_mask_rcnn_x101_64x4d_fpn_1x/epoch_24.pth')
# test a list of images
img_path = '/nfs/project/libo_i/mmdetection/data/kitti/testing/kitti_demo_image'
fs = os.listdir(img_path)
imgs = []
for item in fs:
item_path = os.path.join(img_path, item)
if not os.path.isdir(item_path):
imgs.append(item_path)
for i, result in enumerate(inference_detector(model, imgs, cfg, device='cuda:0')):
print(i, imgs[i])
save_path = "/nfs/project/libo_i/mmdetection/selfmade/kitti_demo_image/24_kitti_inferred/{}".format(os.path.basename(imgs[i]))
print(save_path)
show_result(imgs[i], result, dataset='kitti', outfile=save_path)
def infer_image(self, image):
# compute prediction
result = inference_detector(self.model, image)
# import numpy as np
bbox_result, segm_result = result, None
bboxes = np.vstack(bbox_result)
labels = [
np.full(bbox.shape[0], i, dtype=np.int32)
for i, bbox in enumerate(bbox_result)
]
labels = np.concatenate(labels)
score_thr = 0.3
if score_thr > 0:
assert bboxes.shape[1] == 5
scores = bboxes[:, -1]
inds = scores > score_thr
scores = bboxes[inds, -1]
boxes = bboxes[inds, :4]
labels = labels[inds]
scores = scores.tolist()
boxes = boxes.tolist()
labels = labels.tolist()
# print("scores:", scores)
# print("bboxes:", boxes)
# print("labels", labels)
labels_words = [str(self.classes[idx]) for idx in labels]
# print(labels_words)
return {
'boxes': boxes,
'labels_words': labels_words,
'labels_idx': labels,
'scores': scores
}
def solo_infer(solo_cp_path, src_folder, dst_folder, json_path):
config_file = '../configs/solo/decoupled_solo_r50_fpn_8gpu_3x.py'
# 加载solo模型
model = init_detector(config_file, solo_cp_path, device='cuda:0')
# 创建json文件
fjson = open(json_path, 'w')
# 读取所有输入图像的路径
all_img_paths = glob.glob(os.path.join(src_folder, '*/*.*'))
num_img = len(all_img_paths)
# 开始处理
state = {'num': float(num_img)}
content = [] # 所有图像的信息
for i in tqdm(range(num_img)):
img_path = all_img_paths[i]
img_name = img_path.split('/')[-1]
if cv2.imread(img_path) is None:
print(img_path)
continue
result = inference_detector(model, img_path) # 单张图像通过solo
dst_path = os.path.join(dst_folder, img_name) # 以相同图像名称保存
show_result_ins(img_path,
result,
model.CLASSES,
score_thr=0.25,
out_file=dst_path) # 绘制 mask并保存
# 分析mask信息
img_info = post_treatment(img_path,
result,
model.CLASSES,
score_thr=0.25)
# print(img_info)
content.append(img_info)
# 保存到json文件
state['content'] = content
json.dump(state, fjson, indent=4)
fjson.close()
def predict(self, tasks, **kwargs):
assert len(tasks) == 1
task = tasks[0]
image_path = get_image_local_path(task["data"][self.value])
model_results = inference_detector(self.model, image_path)
results = []
all_scores = []
img_width, img_height = get_image_size(image_path)
for bboxes, label in zip(model_results, self.model.CLASSES):
output_label = self.label_map.get(label, label)
if output_label not in self.labels_in_config:
print(output_label + " label not found in project config.")
continue
for bbox in bboxes:
bbox = list(bbox)
if not bbox:
continue
score = float(bbox[-1])
if score < self.score_thresh:
continue
x, y, xmax, ymax = bbox[:4]
results.append({
"from_name": self.from_name,
"to_name": self.to_name,
"type": "rectanglelabels",
"value": {
"rectanglelabels": [output_label],
"x": int(x / img_width * 100),
"y": int(y / img_height * 100),
"width": int((xmax - x) / img_width * 100),
"height": int((ymax - y) / img_height * 100),
},
"score": score,
})
all_scores.append(score)
avg_score = sum(all_scores) / len(all_scores)
return [{"result": results, "score": avg_score}]
def inference_split(self,
img: Path,
threshold: float = DEFAULT_THRESHOLD,
padding: int = 0):
image = cv2.imread(str(img.absolute()))
split_weight = [0.2, 0.43, 0.65, 0.84, 1]
splits = []
prev = 0
for split_w in split_weight:
border = int(split_w * image.shape[0])
splits.append(image[prev:border])
prev = border
tables = []
headers = []
prev_shape = 0
for num, split in enumerate(splits):
split = add_padding(split, padding)
result = inference_detector(self.model, split)
if self.should_visualize:
inference_image = self.model.show_result(split,
result,
thickness=2)
image_path = (img.parent.parent / "raw_model" / img.name /
f"{num}.png")
image_path.parent.mkdir(parents=True, exist_ok=True)
cv2.imwrite(str(image_path.absolute()), inference_image)
inf_tables, header, _ = inference_result_to_boxes(
extract_boxes_from_result(result,
CLASS_NAMES,
score_thr=threshold))
crop_padding([(inf_tables, headers)], padding)
shift(inf_tables, headers, prev_shape, 0)
tables.extend(inf_tables)
headers.extend(header)
prev_shape += split.shape[0]
return tables, headers
def inference(self, preprocessed_image):
"""
predict bounding boxes of a preprocessed image
Args:
preprocessed_image: a preprocessed RGB frame
Returns:
A list of dictionaries, each item has the id, relative bounding box coordinate and prediction confidence score
of one detected instance.
"""
self.frame = preprocessed_image
output_dict = inference_detector(self.detection_model, preprocessed_image)
class_id = int(self.config.get_section_dict('Teacher')['ClassID'])
score_threshold = float(self.config.get_section_dict('Teacher')['MinScore'])
result = []
for i, box in enumerate(output_dict[0]): # number of boxes
if box[-1] > score_threshold:
result.append({"id": str(class_id) + '-' + str(i),
"bbox": [box[0] / self.image_size[0], box[1] / self.image_size[1],
box[2] / self.image_size[0], box[3] / self.image_size[1]],
"score": box[-1]})
return result
def inference_model(config_name, checkpoint, args, logger=None):
cfg = Config.fromfile(config_name)
if args.aug:
if 'flip' in cfg.data.test.pipeline[1]:
cfg.data.test.pipeline[1].flip = True
else:
if logger is not None:
logger.error(f'{config_name}: unable to start aug test')
else:
print(f'{config_name}: unable to start aug test', flush=True)
model = init_detector(cfg, checkpoint, device=args.device)
# test a single image
result = inference_detector(model, args.img)
# show the results
if args.show:
show_result_pyplot(model,
args.img,
result,
score_thr=args.score_thr,
wait_time=args.wait_time)
return result
def mmdInferenceOnExternalImages(external_input_images_path, output_result_path, mmd_config_file, mmd_checkpoint):
"""Note that, external_input_images_path puts all rgb images instead of both images and depths folder."""
# build the model from a config file and a checkpoint file
model = init_detector(mmd_config_file, mmd_checkpoint, device='cuda:0')
# test a single image and show the results
images = os.listdir(external_input_images_path)
results = {}
with tqdm(total=len(images)) as pbar:
for image_name in images:
image_path = os.path.join(external_input_images_path, image_name)
result = inference_detector(model, image_path)
output_path = os.path.join(output_result_path, image_name)
# visualize the results in a new window
# show_result(image_path, result, model.CLASSES)
# or save the visualization results to image files
show_result(image_path, result, model.CLASSES, show=False, out_file=output_path)
results[image_name] = result
pbar.update(1)
return results
def detect_logo_location(image_path):
list_img =[]
img = mmcv.imread(image_path)
j=0
dst_dir = "/content/drive/My Drive/logo detection: phase 2/dataset for testing/crop"
result = inference_detector(mmdetection_model, img) # lấy bouding box từ hình sử dụng model mmdetection
if result: # nếu tập không rỗng( có tồn tại logo)
img = Image.open(image_path)
for i in range(len(result[0])): # loop qua từng bounding box trong hình
if result[0][i][4] >= 0.3: # nếu probality > 0.5 lấy bouding và crop hình
left = result[0][i][0]
top = result[0][i][1]
right = result[0][i][2]
bottom = result[0][i][3]
im_crop = img.crop((left, top, right, bottom))
if im_crop.mode == "RGBA":
im_crop = im_crop.convert('RGB')
name = os.path.join(dst_dir, "crop_" + str(j) + ".jpg")
im_crop.save(name)
j=j+1
list_img.append(name)
return list_img
def show_image(image_array, model, fps_s, format_='jpeg'):
f = BytesIO()
result = inference_detector(model, image_array)
result[0][:, -1] = 0
img_out = model.show_result(
image_array,
result,
score_thr=0.35,
show=False,
font_scale=0.0,
thickness=3,
bbox_color='green',
)
fps = fps_s[-1] if fps_s else 0.0
cv2.putText(img_out,
str(fps)[:5], (1200, 20), cv2.FONT_HERSHEY_COMPLEX, 0.9,
(0, 255, 0))
img_out = cv2.cvtColor(img_out, cv2.COLOR_BGR2RGB)
pil_img = PIL.Image.fromarray(img_out)
pil_img.save(f, format=format_)
IPython.display.display(IPython.display.Image(data=f.getvalue()))
def detector(self,img):
'''
:param img: opencv读取的图片，底层为numpy格式
:return: result ,size=[n*6]，其中n为检测到的目标数量
result[:,0]=目标类别
result[:,1:4]=x_min,y_min,x_max,y_max
result[:,5]=置信度
'''
self.det_result=inference_detector(self.net_model,img)
labels = [
np.full(bbox.shape[0], i, dtype=np.int32)
for i, bbox in enumerate(self.det_result)
]
self.labels = np.concatenate(labels)
self.labels=self.labels.reshape(self.config.num_roi,1)
self.bboxes = np.vstack(self.det_result)
self.result=np.hstack((self.labels,self.bboxes))
# 基于重叠度的筛选
nms_iou_detec=nms_iou(self.config.miji_thr,self.config.h_thr,self.config.nms_thr,self.config.num_thr)
self.result=nms_iou_detec.mix_nms(self.result)
return self.result
def SlideIndex(PredImg,im_height,im_width,model):
new_name = len(os.listdir(SavePath)) + 1
bboxouts = []
for imgi in tqdm(range(0, im_height,image_size[0])):
for imgj in range(0, im_width,image_size[1]):
cropped = PredImg[imgi:imgi+image_size[0], imgj:imgj+image_size[1],:]
if cropped.shape[0]< image_size[0] or cropped.shape[1]< image_size[1]:
NewI = np.zeros((image_size[0], image_size[0], 3))
NewI[:cropped.shape[0],:cropped.shape[1], :] = cropped[:,:,:]
cropped = NewI
resultbbox = inference_detector(model, cropped)
# # 切片文件保存
# singleImg = visualize(cropped,resultbbox)
# cv2.imwrite(SavePath + "/%d.png"%new_name,singleImg)
# cv2.destroyAllWindows()
# bbox偏移
bboxTrans = bbox_transfrom(imgi,imgj,resultbbox)
bboxouts.append(resultbbox)
new_name=new_name+1
return bboxouts
def main():
args = parse_args()
model = init_detector(args.config,
args.checkpoint,
device=torch.device('cuda', args.device))
camera = cv2.VideoCapture(args.camera_id)
print('Press "Esc", "q" or "Q" to exit.')
while True:
ret_val, img = camera.read()
result = inference_detector(model, img)
ch = cv2.waitKey(1)
if ch == 27 or ch == ord('q') or ch == ord('Q'):
break
show_result(img,
result,
model.CLASSES,
score_thr=args.score_thr,
wait_time=1)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--config', help= 'test config file path')
parser.add_argument('--checkpoint', help= 'checkpoint file')
args = parser.parse_args()
config_file = args.config
checkpoint_file = args.checkpoint
model = init_detector(config_file, checkpoint_file, device= 'cuda:0')
image_path = os.path.abspath("./data/only_carTruck_frame/")
result_path = os.path.abspath("./data/only_carTruck_inference")
images = sorted(os.listdir(image_path))
integer = 0
# images에 있는 image를 하나씩 뽑아가면서 inference 시켜야함
for imgs in images:
img_for_inference = image_path + "/" + imgs
result = inference_detector(model, img_for_inference)
last_path = result_path + "/{0:06d}".format(integer) + ".jpg"
save_result_pyplot(model, img_for_inference, result, last_path)
integer += 1
def main():
args = get_args()
# build the model from a config file and a checkpoint file
model = init_detector(args.config, args.ckpt, device='cuda:0')
if len(args.input) == 1:
args.input = glob.glob(args.input[0])
for img_path in args.input:
# test a single image and show the results
# img = 'test.jpg' # or img = mmcv.imread(img), which will only load it once
img = mmcv.imread(img_path)
result = inference_detector(model, img)
# visualize the results in a new window
out_file = None if args.output is None else osp.join(
args.output,
str(time.time()) + '.jpg')
show_result(img,
result,
model.CLASSES,
score_thr=args.score_thr,
out_file=out_file)
def get_frt_fb(img, img_name):
detections = inference_detector(model_fb, img)
detections = detections[0]
scores = detections[:, 4]
if len(scores) > 0:
max_score = np.max(scores) # 把最大置信度认为赛第四类的概率
cnt = 0
for j in range(len(detections)):
score = scores[j]
if score > 0.7: # 认为大于0.7的肯定大概率是第四类
cnt += 1
if max_score > 0.9: # 大于0.9直接认为是第四类
is_fb = 3
else:
is_fb = -1
else:
max_score = 0
is_fb = -1
cnt = 0
features_dic_fb['name'].append(img_name)
features_dic_fb['fb_cnt'].append(cnt)
features_dic_fb['fb_prob'].append(max_score)
features_dic_fb['is_fb'].append(is_fb)
def detection_inference(args, frame_paths):
"""Detect human boxes given frame paths.
Args:
args (argparse.Namespace): The arguments.
frame_paths (list[str]): The paths of frames to do detection inference.
Returns:
list[np.ndarray]: The human detection results.
"""
model = init_detector(args.det_config, args.det_checkpoint, args.device)
assert model.CLASSES[0] == 'person', ('We require you to use a detector '
'trained on COCO')
results = []
print('Performing Human Detection for each frame')
prog_bar = mmcv.ProgressBar(len(frame_paths))
for frame_path in frame_paths:
result = inference_detector(model, frame_path)
# We only keep human detections with score larger than det_score_thr
result = result[0][result[0][:, 4] >= args.det_score_thr]
results.append(result)
prog_bar.update()
return results
def main():
parser = ArgumentParser()
#parser.add_argument('img', help='Image file')
parser.add_argument('config', help='Config file')
parser.add_argument('checkpoint', help='Checkpoint file')
parser.add_argument(
'--device', default='cuda:0', help='Device used for inference')
parser.add_argument(
'--score-thr', type=float, default=0.5, help='bbox score threshold')
parser.add_argument('--path', type=str, default='/home/ryo/W2020V1_142_NECvsAGE_F8_200119.MP4')
parser.add_argument('--out-dir', type=str, default='./')
args = parser.parse_args()
# build the model from a config file and a checkpoint file
model = init_detector(args.config, args.checkpoint, device=args.device)
cap_file = cv2.VideoCapture(args.path)
out = cv2.VideoWriter('outpy.avi',cv2.VideoWriter_fourcc('M','J','P','G'), 30, (1280,720))
count = 0
while (cap_file.isOpened()):
print(count)
if count > 10000:
break
count+=1
ret, frame = cap_file.read()
if ret:
result = inference_detector(model, frame)
result = np.vstack(result[0])
bboxes = result[result[:, -1] > args.score_thr]
for j in range(bboxes.shape[0]):
left_top = (bboxes[j, 0], bboxes[j, 1])
right_bottom = (bboxes[j, 2], bboxes[j, 3])
cv2.rectangle(frame, left_top, right_bottom, (0, 0 ,255), thickness=1)
out.write(frame)
else:
break
cap_file.release()
out.release()
def run(self):
# 测试一张图片
dirname = self.dirname
imgs = os.listdir(dirname)
if imgs is not None:
imgs = sorted(imgs, key=lambda x: int(x.split('/')[-1].split('.')[0]))
img_size = cv2.imread(dirname + "/" + imgs[0]).shape
flag = 0
for img in imgs:
print(img)
path = dirname + "/" + img
id = img.split('.')[0]
result = inference_detector(self.model, path)
if isinstance(result, tuple):
bbox_result, segm_result = result
else:
bbox_result, segm_result = result, None
bboxes = np.vstack(bbox_result)
scores = bboxes[:, -1]
inds = scores > self.score
bboxes = bboxes[inds, :]
temp = []
for bbox in bboxes:
bbox_int = bbox.astype(np.int32)
temp.append(bbox_int)
newbbox = self.check_bbox(temp)
filename = id + ".tif"
xmlwriter = PascalVocWriter("VOC2007", filename, img_size)
for box in newbbox:
xmlwriter.addBndBox(box[0], box[1], box[2], box[3], "nano", "0")
xmlwriter.save(self.defaultSaveDir + "/" + id + ".xml")
flag += 1
if self.quick_flag == 1:
break
prograssbar_value = round(flag / len(imgs), 2) * 100
self.progressBarValue[int].emit(int(prograssbar_value))
def process(self, img) -> ShelfInfo:
'''
Process the image and get the segmentation result for the every row
:param img: numpy array image
:return: the ShelfInfo
'''
prediction = inference_detector(self.model, img)
polygon_fit = PolygonFit()
results = []
for bbox, seg in zip(prediction[0][0], prediction[1][0]):
if bbox[-1] < self.score_thr:
continue
contours, hierarchy = cv2.findContours(np.array(seg, dtype=np.uint8), cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
contours = [cv2.convexHull(c) for c in contours]
contours = [cv2.approxPolyDP(c, 0.001 * cv2.arcLength(c, True), True) for c in contours]
contours = [c for c in contours if c.shape[0] > 2]
if len(contours) == 0:
continue
elif len(contours) > 1:
contour_areas = [cv2.contourArea(c) for c in contours]
max_idx = np.argmax(contour_areas)
contour = contours[max_idx]
else:
contour = contours[0]
contour = contour.reshape(-1, 2)
polygon = polygon_fit.fit(contour)
results.append(dict(
bbox=bbox[:-1].round().astype(np.int32).tolist(),
points=contour.tolist(),
polygon=polygon,
score=float(bbox[-1])
))
results.sort(key=lambda x: x['bbox'][1])
results = remove_overlap(results)
results = filter_results(results)
return results
def predict():
CONFIDENCE_THRESHOLD = 0.0001
args = parse_args()
cfg = mmcv.Config.fromfile(args.cfg)
cfg.model.pretrained = None
# construct the model and load checkpoint
#model = build_detector(cfg.model, test_cfg=cfg.test_cfg)
model = init_detector(args.cfg, args.weights, device='cuda:0')
#_ = load_checkpoint(model, 'https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/faster_rcnn_r50_fpn_1x_20181010-3d1b3351.pth')
#for image in tqdm(glob("/mfs/home/limengwei/car_face/car_face/object_detection_data_back_seats/*/*.png")):
for image in tqdm(glob("/mfs/home/limengwei/car_face/car_face/object_detection_data_angle_top_head/images_val/*.png")):
#for image in tqdm(glob("/mfs/home/limengwei/car_face/car_face/object_detection_data_both_side_back_seats/*/*.png")):
img = mmcv.imread(image)
filename = image.split("/")[-1]
start = time.time()
result = inference_detector(model, img)
print(time.time()-start)
labels = np.concatenate([
np.full(bbox.shape[0], i, dtype=np.int32)
for i, bbox in enumerate(result)])
#labels = labels + 1
bboxes = np.vstack(result)
for label,bbox in zip(labels,bboxes):
threshold = bbox[-1]
if threshold < CONFIDENCE_THRESHOLD:
continue
x1,y1 = bbox[0],bbox[1]
x2,y2 = bbox[2],bbox[1]
x3,y3 = bbox[2],bbox[3]
x4,y4 = bbox[0],bbox[3]
text = id_to_label[label+1]
cv2.putText(img,text,(int(x1),int(y1-2)),cv2.FONT_HERSHEY_COMPLEX,0.8,(0,255,255))
cv2.rectangle(img,(int(bbox[0]),int(bbox[1])),(int(bbox[2]),int(bbox[3])),(0,0,255),2)
status = cv2.imwrite("outputs/%s"%filename,img)
#embed()
cv2.imshow("img",img)
cv2.waitKey(0)