Aniaaaa/code_site
0
1# YOLOv5 🚀 by Ultralytics, GPL-3.0 license2"""3Run inference on images, videos, directories, streams, etc.4 5Usage - sources:6 $ python path/to/detect.py --weights yolov5s.pt --source 0 # webcam7 img.jpg # image8 vid.mp4 # video9 path/ # directory10 path/*.jpg # glob11 'https://youtu.be/Zgi9g1ksQHc' # YouTube12 'rtsp://example.com/media.mp4' # RTSP, RTMP, HTTP stream13 14Usage - formats:15 $ python path/to/detect.py --weights yolov5s.pt # PyTorch16 yolov5s.torchscript # TorchScript17 yolov5s.onnx # ONNX Runtime or OpenCV DNN with --dnn18 yolov5s.xml # OpenVINO19 yolov5s.engine # TensorRT20 yolov5s.mlmodel # CoreML (macOS-only)21 yolov5s_saved_model # TensorFlow SavedModel22 yolov5s.pb # TensorFlow GraphDef23 yolov5s.tflite # TensorFlow Lite24 yolov5s_edgetpu.tflite # TensorFlow Edge TPU25"""26 27import argparse28import os29import sys30from pathlib import Path31 32import torch33import torch.backends.cudnn as cudnn34 35FILE = Path(__file__).resolve()36ROOT = FILE.parents[0] # YOLOv5 root directory37if str(ROOT) not in sys.path:38 sys.path.append(str(ROOT)) # add ROOT to PATH39ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative40from models.common import DetectMultiBackend41from utils.dataloaders import IMG_FORMATS, VID_FORMATS, LoadImages, LoadStreams42from utils.general import (LOGGER, check_file, check_img_size, check_imshow, check_requirements, colorstr, cv2,43 increment_path, non_max_suppression, print_args, scale_coords, strip_optimizer, xyxy2xywh)44from utils.plots import Annotator, colors, save_one_box45from utils.torch_utils import select_device, time_sync46 47 48@torch.no_grad()49def run(50 weights=ROOT / 'yolov5s.pt', # model.pt path(s)51 source=ROOT / 'data/images', # file/dir/URL/glob, 0 for webcam52 data=ROOT / 'data/coco128.yaml', # dataset.yaml path53 imgsz=(640, 640), # inference size (height, width)54 conf_thres=0.25, # confidence threshold55 iou_thres=0.45, # NMS IOU threshold56 max_det=1000, # maximum detections per image57 device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu58 view_img=False, # show results59 save_txt=False, # save results to *.txt60 save_conf=False, # save confidences in --save-txt labels61 save_crop=False, # save cropped prediction boxes62 nosave=False, # do not save images/videos63 classes=None, # filter by class: --class 0, or --class 0 2 364 agnostic_nms=False, # class-agnostic NMS65 augment=False, # augmented inference66 visualize=False, # visualize features67 update=False, # update all models68 project=ROOT / 'runs/detect', # save results to project/name69 name='exp', # save results to project/name70 exist_ok=False, # existing project/name ok, do not increment71 line_thickness=3, # bounding box thickness (pixels)72 hide_labels=False, # hide labels73 hide_conf=False, # hide confidences74 half=False, # use FP16 half-precision inference75 dnn=False, # use OpenCV DNN for ONNX inference76):77 source = str(source)78 save_img = not nosave and not source.endswith('.txt') # save inference images79 is_file = Path(source).suffix[1:] in (IMG_FORMATS + VID_FORMATS)80 is_url = source.lower().startswith(('rtsp://', 'rtmp://', 'http://', 'https://'))81 webcam = source.isnumeric() or source.endswith('.txt') or (is_url and not is_file)82 if is_url and is_file:83 source = check_file(source) # download84 85 # Directories86 save_dir = increment_path(Path(project) / name, exist_ok=exist_ok) # increment run87 (save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir88 89 # Load model90 device = select_device(device)91 print(f'weights : {weights}, device : {device}')92 model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data, fp16=half)93 stride, names, pt = model.stride, model.names, model.pt94 imgsz = check_img_size(imgsz, s=stride) # check image sizerun95 96 # Dataloader97 if webcam:98 view_img = check_imshow()99 cudnn.benchmark = True # set True to speed up constant image size inference100 dataset = LoadStreams(source, img_size=imgsz, stride=stride, auto=pt)101 bs = len(dataset) # batch_size102 else:103 dataset = LoadImages(source, img_size=imgsz, stride=stride, auto=pt)104 bs = 1 # batch_size105 vid_path, vid_writer = [None] * bs, [None] * bs106 107 # Run inference108 model.warmup(imgsz=(1 if pt else bs, 3, *imgsz)) # warmup109 seen, windows, dt = 0, [], [0.0, 0.0, 0.0]110 for path, im, im0s, vid_cap, s in dataset:111 t1 = time_sync()112 im = torch.from_numpy(im).to(device)113 im = im.half() if model.fp16 else im.float() # uint8 to fp16/32114 im /= 255 # 0 - 255 to 0.0 - 1.0115 if len(im.shape) == 3:116 im = im[None] # expand for batch dim117 t2 = time_sync()118 dt[0] += t2 - t1119 120 # Inference121 visualize = increment_path(save_dir / Path(path).stem, mkdir=True) if visualize else False122 pred = model(im, augment=augment, visualize=visualize)123 t3 = time_sync()124 dt[1] += t3 - t2125 126 # NMS127 pred = non_max_suppression(pred, conf_thres, iou_thres, classes, agnostic_nms, max_det=max_det)128 dt[2] += time_sync() - t3129 130 # Second-stage classifier (optional)131 # pred = utils.general.apply_classifier(pred, classifier_model, im, im0s)132 133 # Process predictions134 for i, det in enumerate(pred): # per image135 seen += 1136 if webcam: # batch_size >= 1137 p, im0, frame = path[i], im0s[i].copy(), dataset.count138 s += f'{i}: '139 else:140 p, im0, frame = path, im0s.copy(), getattr(dataset, 'frame', 0)141 142 p = Path(p) # to Path143 save_path = str(save_dir / p.name) # im.jpg144 txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}') # im.txt145 s += '%gx%g ' % im.shape[2:] # print string146 gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh147 imc = im0.copy() if save_crop else im0 # for save_crop148 annotator = Annotator(im0, line_width=line_thickness, example=str(names))149 if len(det):150 # Rescale boxes from img_size to im0 size151 det[:, :4] = scale_coords(im.shape[2:], det[:, :4], im0.shape).round()152 153 # Print results154 for c in det[:, -1].unique():155 n = (det[:, -1] == c).sum() # detections per class156 s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string157 158 # Write results159 for *xyxy, conf, cls in reversed(det):160 if save_txt: # Write to file161 xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh162 line = (cls, *xywh, conf) if save_conf else (cls, *xywh) # label format163 with open(f'{txt_path}.txt', 'a') as f:164 f.write(('%g ' * len(line)).rstrip() % line + '\n')165 166 if save_img or save_crop or view_img: # Add bbox to image167 c = int(cls) # integer class168 label = None if hide_labels else (names[c] if hide_conf else f'{names[c]} {conf:.2f}')169 annotator.box_label(xyxy, label, color=colors(c, True))170 if save_crop:171 save_one_box(xyxy, imc, file=save_dir / 'crops' / names[c] / f'{p.stem}.jpg', BGR=True)172 173 # Stream results174 im0 = annotator.result()175 if view_img:176 if p not in windows:177 windows.append(p)178 cv2.namedWindow(str(p), cv2.WINDOW_NORMAL | cv2.WINDOW_KEEPRATIO) # allow window resize (Linux)179 cv2.resizeWindow(str(p), im0.shape[1], im0.shape[0])180 cv2.imshow(str(p), im0)181 cv2.waitKey(1) # 1 millisecond182 183 # Save results (image with detections)184 if save_img:185 if dataset.mode == 'image':186 cv2.imwrite(save_path, im0)187 else: # 'video' or 'stream'188 if vid_path[i] != save_path: # new video189 vid_path[i] = save_path190 if isinstance(vid_writer[i], cv2.VideoWriter):191 vid_writer[i].release() # release previous video writer192 if vid_cap: # video193 fps = vid_cap.get(cv2.CAP_PROP_FPS)194 w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))195 h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))196 else: # stream197 fps, w, h = 30, im0.shape[1], im0.shape[0]198 save_path = str(Path(save_path).with_suffix('.mp4')) # force *.mp4 suffix on results videos199 vid_writer[i] = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h))200 vid_writer[i].write(im0)201 202 # Print time (inference-only)203 LOGGER.info(f'{s}Done. ({t3 - t2:.3f}s)')204 205 # Print results206 t = tuple(x / seen * 1E3 for x in dt) # speeds per image207 LOGGER.info(f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {(1, 3, *imgsz)}' % t)208 if save_txt or save_img:209 s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''210 LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")211 if update:212 strip_optimizer(weights) # update model (to fix SourceChangeWarning)213 214 215def parse_opt():216 parser = argparse.ArgumentParser()217 parser.add_argument('--weights', nargs='+', type=str, default=ROOT / 'yolov5s.pt', help='model path(s)')218 parser.add_argument('--source', type=str, default=ROOT / 'data/images', help='file/dir/URL/glob, 0 for webcam')219 parser.add_argument('--data', type=str, default=ROOT / 'data/coco128.yaml', help='(optional) dataset.yaml path')220 parser.add_argument('--imgsz', '--img', '--img-size', nargs='+', type=int, default=[640], help='inference size h,w')221 parser.add_argument('--conf-thres', type=float, default=0.25, help='confidence threshold')222 parser.add_argument('--iou-thres', type=float, default=0.45, help='NMS IoU threshold')223 parser.add_argument('--max-det', type=int, default=1000, help='maximum detections per image')224 parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')225 parser.add_argument('--view-img', action='store_true', help='show results')226 parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')227 parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')228 parser.add_argument('--save-crop', action='store_true', help='save cropped prediction boxes')229 parser.add_argument('--nosave', action='store_true', help='do not save images/videos')230 parser.add_argument('--classes', nargs='+', type=int, default=(0, 1, 2, 3, 5, 7, 9, 11), help='filter by class: --classes 0, or --classes 0 2 3')231 parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS')232 parser.add_argument('--augment', action='store_true', help='augmented inference')233 parser.add_argument('--visualize', action='store_true', help='visualize features')234 parser.add_argument('--update', action='store_true', help='update all models')235 parser.add_argument('--project', default=ROOT / 'runs/detect', help='save results to project/name')236 parser.add_argument('--name', default='exp', help='save results to project/name')237 parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')238 parser.add_argument('--line-thickness', default=3, type=int, help='bounding box thickness (pixels)')239 parser.add_argument('--hide-labels', default=False, action='store_true', help='hide labels')240 parser.add_argument('--hide-conf', default=False, action='store_true', help='hide confidences')241 parser.add_argument('--half', action='store_true', help='use FP16 half-precision inference')242 parser.add_argument('--dnn', action='store_true', help='use OpenCV DNN for ONNX inference')243 opt = parser.parse_args()244 opt.imgsz *= 2 if len(opt.imgsz) == 1 else 1 # expand245 print_args(vars(opt))246 247 return opt248 249 250def main(opt):251 print(f"wtf opt {opt}")252 check_requirements(exclude=('tensorboard', 'thop'))253 run(**vars(opt))254 255 256if __name__ == "__main__":257 opt = parse_opt()258 print(f"wtf opt {opt}")259 main(opt)260