Ви не можете вибрати більше 25 тем Теми мають розпочинатися з літери або цифри, можуть містити дефіси (-) і не повинні перевищувати 35 символів.

252 lines
13KB

  1. # YOLOv5 🚀 by Ultralytics, GPL-3.0 license
  2. """
  3. Run inference on images, videos, directories, streams, etc.
  4. Usage - sources:
  5. $ python path/to/detect.py --weights yolov5s.pt --source 0 # webcam
  6. img.jpg # image
  7. vid.mp4 # video
  8. path/ # directory
  9. path/*.jpg # glob
  10. 'https://youtu.be/Zgi9g1ksQHc' # YouTube
  11. 'rtsp://example.com/media.mp4' # RTSP, RTMP, HTTP stream
  12. Usage - formats:
  13. $ python path/to/detect.py --weights yolov5s.pt # PyTorch
  14. yolov5s.torchscript # TorchScript
  15. yolov5s.onnx # ONNX Runtime or OpenCV DNN with --dnn
  16. yolov5s.xml # OpenVINO
  17. yolov5s.engine # TensorRT
  18. yolov5s.mlmodel # CoreML (MacOS-only)
  19. yolov5s_saved_model # TensorFlow SavedModel
  20. yolov5s.pb # TensorFlow GraphDef
  21. yolov5s.tflite # TensorFlow Lite
  22. yolov5s_edgetpu.tflite # TensorFlow Edge TPU
  23. """
  24. import argparse
  25. import os
  26. import sys
  27. from pathlib import Path
  28. import torch
  29. import torch.backends.cudnn as cudnn
  30. FILE = Path(__file__).resolve()
  31. ROOT = FILE.parents[0] # YOLOv5 root directory
  32. if str(ROOT) not in sys.path:
  33. sys.path.append(str(ROOT)) # add ROOT to PATH
  34. ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
  35. from models.common import DetectMultiBackend
  36. from utils.datasets import IMG_FORMATS, VID_FORMATS, LoadImages, LoadStreams
  37. from utils.general import (LOGGER, check_file, check_img_size, check_imshow, check_requirements, colorstr, cv2,
  38. increment_path, non_max_suppression, print_args, scale_coords, strip_optimizer, xyxy2xywh)
  39. from utils.plots import Annotator, colors, save_one_box
  40. from utils.torch_utils import select_device, time_sync
  41. @torch.no_grad()
  42. def run(weights=ROOT / 'yolov5s.pt', # model.pt path(s)
  43. source=ROOT / 'data/images', # file/dir/URL/glob, 0 for webcam
  44. data=ROOT / 'data/coco128.yaml', # dataset.yaml path
  45. imgsz=(640, 640), # inference size (height, width)
  46. conf_thres=0.25, # confidence threshold
  47. iou_thres=0.45, # NMS IOU threshold
  48. max_det=1000, # maximum detections per image
  49. device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu
  50. view_img=False, # show results
  51. save_txt=False, # save results to *.txt
  52. save_conf=False, # save confidences in --save-txt labels
  53. save_crop=False, # save cropped prediction boxes
  54. nosave=False, # do not save images/videos
  55. classes=None, # filter by class: --class 0, or --class 0 2 3
  56. agnostic_nms=False, # class-agnostic NMS
  57. augment=False, # augmented inference
  58. visualize=False, # visualize features
  59. update=False, # update all models
  60. project=ROOT / 'runs/detect', # save results to project/name
  61. name='exp', # save results to project/name
  62. exist_ok=False, # existing project/name ok, do not increment
  63. line_thickness=3, # bounding box thickness (pixels)
  64. hide_labels=False, # hide labels
  65. hide_conf=False, # hide confidences
  66. half=False, # use FP16 half-precision inference
  67. dnn=False, # use OpenCV DNN for ONNX inference
  68. ):
  69. source = str(source)
  70. save_img = not nosave and not source.endswith('.txt') # save inference images
  71. is_file = Path(source).suffix[1:] in (IMG_FORMATS + VID_FORMATS)
  72. is_url = source.lower().startswith(('rtsp://', 'rtmp://', 'http://', 'https://'))
  73. webcam = source.isnumeric() or source.endswith('.txt') or (is_url and not is_file)
  74. if is_url and is_file:
  75. source = check_file(source) # download
  76. # Directories
  77. save_dir = increment_path(Path(project) / name, exist_ok=exist_ok) # increment run
  78. (save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
  79. # Load model
  80. device = select_device(device)
  81. model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data, fp16=half)
  82. stride, names, pt = model.stride, model.names, model.pt
  83. imgsz = check_img_size(imgsz, s=stride) # check image size
  84. # Dataloader
  85. if webcam:
  86. view_img = check_imshow()
  87. cudnn.benchmark = True # set True to speed up constant image size inference
  88. dataset = LoadStreams(source, img_size=imgsz, stride=stride, auto=pt)
  89. bs = len(dataset) # batch_size
  90. else:
  91. dataset = LoadImages(source, img_size=imgsz, stride=stride, auto=pt)
  92. bs = 1 # batch_size
  93. vid_path, vid_writer = [None] * bs, [None] * bs
  94. # Run inference
  95. model.warmup(imgsz=(1 if pt else bs, 3, *imgsz)) # warmup
  96. dt, seen = [0.0, 0.0, 0.0], 0
  97. for path, im, im0s, vid_cap, s in dataset:
  98. t1 = time_sync()
  99. im = torch.from_numpy(im).to(device)
  100. im = im.half() if model.fp16 else im.float() # uint8 to fp16/32
  101. im /= 255 # 0 - 255 to 0.0 - 1.0
  102. if len(im.shape) == 3:
  103. im = im[None] # expand for batch dim
  104. t2 = time_sync()
  105. dt[0] += t2 - t1
  106. # Inference
  107. visualize = increment_path(save_dir / Path(path).stem, mkdir=True) if visualize else False
  108. pred = model(im, augment=augment, visualize=visualize)
  109. t3 = time_sync()
  110. dt[1] += t3 - t2
  111. # NMS
  112. pred = non_max_suppression(pred, conf_thres, iou_thres, classes, agnostic_nms, max_det=max_det)
  113. dt[2] += time_sync() - t3
  114. # Second-stage classifier (optional)
  115. # pred = utils.general.apply_classifier(pred, classifier_model, im, im0s)
  116. # Process predictions
  117. for i, det in enumerate(pred): # per image
  118. seen += 1
  119. if webcam: # batch_size >= 1
  120. p, im0, frame = path[i], im0s[i].copy(), dataset.count
  121. s += f'{i}: '
  122. else:
  123. p, im0, frame = path, im0s.copy(), getattr(dataset, 'frame', 0)
  124. p = Path(p) # to Path
  125. save_path = str(save_dir / p.name) # im.jpg
  126. txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}') # im.txt
  127. s += '%gx%g ' % im.shape[2:] # print string
  128. gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
  129. imc = im0.copy() if save_crop else im0 # for save_crop
  130. annotator = Annotator(im0, line_width=line_thickness, example=str(names))
  131. if len(det):
  132. # Rescale boxes from img_size to im0 size
  133. det[:, :4] = scale_coords(im.shape[2:], det[:, :4], im0.shape).round()
  134. # Print results
  135. for c in det[:, -1].unique():
  136. n = (det[:, -1] == c).sum() # detections per class
  137. s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
  138. # Write results
  139. for *xyxy, conf, cls in reversed(det):
  140. if save_txt: # Write to file
  141. xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
  142. line = (cls, *xywh, conf) if save_conf else (cls, *xywh) # label format
  143. with open(txt_path + '.txt', 'a') as f:
  144. f.write(('%g ' * len(line)).rstrip() % line + '\n')
  145. if save_img or save_crop or view_img: # Add bbox to image
  146. c = int(cls) # integer class
  147. label = None if hide_labels else (names[c] if hide_conf else f'{names[c]} {conf:.2f}')
  148. annotator.box_label(xyxy, label, color=colors(c, True))
  149. if save_crop:
  150. save_one_box(xyxy, imc, file=save_dir / 'crops' / names[c] / f'{p.stem}.jpg', BGR=True)
  151. # Stream results
  152. im0 = annotator.result()
  153. if view_img:
  154. cv2.imshow(str(p), im0)
  155. cv2.waitKey(1) # 1 millisecond
  156. # Save results (image with detections)
  157. if save_img:
  158. if dataset.mode == 'image':
  159. cv2.imwrite(save_path, im0)
  160. else: # 'video' or 'stream'
  161. if vid_path[i] != save_path: # new video
  162. vid_path[i] = save_path
  163. if isinstance(vid_writer[i], cv2.VideoWriter):
  164. vid_writer[i].release() # release previous video writer
  165. if vid_cap: # video
  166. fps = vid_cap.get(cv2.CAP_PROP_FPS)
  167. w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
  168. h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
  169. else: # stream
  170. fps, w, h = 30, im0.shape[1], im0.shape[0]
  171. save_path = str(Path(save_path).with_suffix('.mp4')) # force *.mp4 suffix on results videos
  172. vid_writer[i] = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h))
  173. vid_writer[i].write(im0)
  174. # Print time (inference-only)
  175. LOGGER.info(f'{s}Done. ({t3 - t2:.3f}s)')
  176. # Print results
  177. t = tuple(x / seen * 1E3 for x in dt) # speeds per image
  178. LOGGER.info(f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {(1, 3, *imgsz)}' % t)
  179. if save_txt or save_img:
  180. s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
  181. LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")
  182. if update:
  183. strip_optimizer(weights) # update model (to fix SourceChangeWarning)
  184. def parse_opt():
  185. parser = argparse.ArgumentParser()
  186. parser.add_argument('--weights', nargs='+', type=str, default=ROOT / 'yolov5s.pt', help='model path(s)')
  187. parser.add_argument('--source', type=str, default=ROOT / 'data/images', help='file/dir/URL/glob, 0 for webcam')
  188. parser.add_argument('--data', type=str, default=ROOT / 'data/coco128.yaml', help='(optional) dataset.yaml path')
  189. parser.add_argument('--imgsz', '--img', '--img-size', nargs='+', type=int, default=[640], help='inference size h,w')
  190. parser.add_argument('--conf-thres', type=float, default=0.25, help='confidence threshold')
  191. parser.add_argument('--iou-thres', type=float, default=0.45, help='NMS IoU threshold')
  192. parser.add_argument('--max-det', type=int, default=1000, help='maximum detections per image')
  193. parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
  194. parser.add_argument('--view-img', action='store_true', help='show results')
  195. parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
  196. parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
  197. parser.add_argument('--save-crop', action='store_true', help='save cropped prediction boxes')
  198. parser.add_argument('--nosave', action='store_true', help='do not save images/videos')
  199. parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --classes 0, or --classes 0 2 3')
  200. parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS')
  201. parser.add_argument('--augment', action='store_true', help='augmented inference')
  202. parser.add_argument('--visualize', action='store_true', help='visualize features')
  203. parser.add_argument('--update', action='store_true', help='update all models')
  204. parser.add_argument('--project', default=ROOT / 'runs/detect', help='save results to project/name')
  205. parser.add_argument('--name', default='exp', help='save results to project/name')
  206. parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
  207. parser.add_argument('--line-thickness', default=3, type=int, help='bounding box thickness (pixels)')
  208. parser.add_argument('--hide-labels', default=False, action='store_true', help='hide labels')
  209. parser.add_argument('--hide-conf', default=False, action='store_true', help='hide confidences')
  210. parser.add_argument('--half', action='store_true', help='use FP16 half-precision inference')
  211. parser.add_argument('--dnn', action='store_true', help='use OpenCV DNN for ONNX inference')
  212. opt = parser.parse_args()
  213. opt.imgsz *= 2 if len(opt.imgsz) == 1 else 1 # expand
  214. print_args(FILE.stem, opt)
  215. return opt
  216. def main(opt):
  217. check_requirements(exclude=('tensorboard', 'thop'))
  218. run(**vars(opt))
  219. if __name__ == "__main__":
  220. opt = parse_opt()
  221. main(opt)