You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

detect.py 9.2KB

4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184
  1. import argparse
  2. import time
  3. from pathlib import Path
  4. import cv2
  5. import torch
  6. import torch.backends.cudnn as cudnn
  7. from models.experimental import attempt_load
  8. from utils.datasets import LoadStreams, LoadImages
  9. from utils.general import check_img_size, check_requirements, check_imshow, non_max_suppression, apply_classifier, \
  10. scale_coords, xyxy2xywh, strip_optimizer, set_logging, increment_path, save_one_box
  11. from utils.plots import colors, plot_one_box
  12. from utils.torch_utils import select_device, load_classifier, time_synchronized
  13. def detect(opt):
  14. source, weights, view_img, save_txt, imgsz = opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
  15. save_img = not opt.nosave and not source.endswith('.txt') # save inference images
  16. webcam = source.isnumeric() or source.endswith('.txt') or source.lower().startswith(
  17. ('rtsp://', 'rtmp://', 'http://', 'https://'))
  18. # Directories
  19. save_dir = increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok) # increment run
  20. (save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
  21. # Initialize
  22. set_logging()
  23. device = select_device(opt.device)
  24. half = device.type != 'cpu' # half precision only supported on CUDA
  25. # Load model
  26. model = attempt_load(weights, map_location=device) # load FP32 model
  27. stride = int(model.stride.max()) # model stride
  28. imgsz = check_img_size(imgsz, s=stride) # check img_size
  29. names = model.module.names if hasattr(model, 'module') else model.names # get class names
  30. if half:
  31. model.half() # to FP16
  32. # Second-stage classifier
  33. classify = False
  34. if classify:
  35. modelc = load_classifier(name='resnet101', n=2) # initialize
  36. modelc.load_state_dict(torch.load('weights/resnet101.pt', map_location=device)['model']).to(device).eval()
  37. # Set Dataloader
  38. vid_path, vid_writer = None, None
  39. if webcam:
  40. view_img = check_imshow()
  41. cudnn.benchmark = True # set True to speed up constant image size inference
  42. dataset = LoadStreams(source, img_size=imgsz, stride=stride)
  43. else:
  44. dataset = LoadImages(source, img_size=imgsz, stride=stride)
  45. # Run inference
  46. if device.type != 'cpu':
  47. model(torch.zeros(1, 3, imgsz, imgsz).to(device).type_as(next(model.parameters()))) # run once
  48. t0 = time.time()
  49. for path, img, im0s, vid_cap in dataset:
  50. img = torch.from_numpy(img).to(device)
  51. img = img.half() if half else img.float() # uint8 to fp16/32
  52. img /= 255.0 # 0 - 255 to 0.0 - 1.0
  53. if img.ndimension() == 3:
  54. img = img.unsqueeze(0)
  55. # Inference
  56. t1 = time_synchronized()
  57. pred = model(img, augment=opt.augment)[0]
  58. # Apply NMS
  59. pred = non_max_suppression(pred, opt.conf_thres, opt.iou_thres, opt.classes, opt.agnostic_nms,
  60. max_det=opt.max_det)
  61. t2 = time_synchronized()
  62. # Apply Classifier
  63. if classify:
  64. pred = apply_classifier(pred, modelc, img, im0s)
  65. # Process detections
  66. for i, det in enumerate(pred): # detections per image
  67. if webcam: # batch_size >= 1
  68. p, s, im0, frame = path[i], f'{i}: ', im0s[i].copy(), dataset.count
  69. else:
  70. p, s, im0, frame = path, '', im0s.copy(), getattr(dataset, 'frame', 0)
  71. p = Path(p) # to Path
  72. save_path = str(save_dir / p.name) # img.jpg
  73. txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}') # img.txt
  74. s += '%gx%g ' % img.shape[2:] # print string
  75. gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
  76. imc = im0.copy() if opt.save_crop else im0 # for opt.save_crop
  77. if len(det):
  78. # Rescale boxes from img_size to im0 size
  79. det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round()
  80. # Print results
  81. for c in det[:, -1].unique():
  82. n = (det[:, -1] == c).sum() # detections per class
  83. s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
  84. # Write results
  85. for *xyxy, conf, cls in reversed(det):
  86. if save_txt: # Write to file
  87. xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
  88. line = (cls, *xywh, conf) if opt.save_conf else (cls, *xywh) # label format
  89. with open(txt_path + '.txt', 'a') as f:
  90. f.write(('%g ' * len(line)).rstrip() % line + '\n')
  91. if save_img or opt.save_crop or view_img: # Add bbox to image
  92. c = int(cls) # integer class
  93. label = None if opt.hide_labels else (names[c] if opt.hide_conf else f'{names[c]} {conf:.2f}')
  94. plot_one_box(xyxy, im0, label=label, color=colors(c, True), line_thickness=opt.line_thickness)
  95. if opt.save_crop:
  96. save_one_box(xyxy, imc, file=save_dir / 'crops' / names[c] / f'{p.stem}.jpg', BGR=True)
  97. # Print time (inference + NMS)
  98. print(f'{s}Done. ({t2 - t1:.3f}s)')
  99. # Stream results
  100. if view_img:
  101. cv2.imshow(str(p), im0)
  102. cv2.waitKey(1) # 1 millisecond
  103. # Save results (image with detections)
  104. if save_img:
  105. if dataset.mode == 'image':
  106. cv2.imwrite(save_path, im0)
  107. else: # 'video' or 'stream'
  108. if vid_path != save_path: # new video
  109. vid_path = save_path
  110. if isinstance(vid_writer, cv2.VideoWriter):
  111. vid_writer.release() # release previous video writer
  112. if vid_cap: # video
  113. fps = vid_cap.get(cv2.CAP_PROP_FPS)
  114. w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
  115. h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
  116. else: # stream
  117. fps, w, h = 30, im0.shape[1], im0.shape[0]
  118. save_path += '.mp4'
  119. vid_writer = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h))
  120. vid_writer.write(im0)
  121. if save_txt or save_img:
  122. s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
  123. print(f"Results saved to {save_dir}{s}")
  124. print(f'Done. ({time.time() - t0:.3f}s)')
  125. if __name__ == '__main__':
  126. parser = argparse.ArgumentParser()
  127. parser.add_argument('--weights', nargs='+', type=str, default='yolov5s.pt', help='model.pt path(s)')
  128. parser.add_argument('--source', type=str, default='data/images', help='source') # file/folder, 0 for webcam
  129. parser.add_argument('--img-size', type=int, default=640, help='inference size (pixels)')
  130. parser.add_argument('--conf-thres', type=float, default=0.25, help='object confidence threshold')
  131. parser.add_argument('--iou-thres', type=float, default=0.45, help='IOU threshold for NMS')
  132. parser.add_argument('--max-det', type=int, default=1000, help='maximum number of detections per image')
  133. parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
  134. parser.add_argument('--view-img', action='store_true', help='display results')
  135. parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
  136. parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
  137. parser.add_argument('--save-crop', action='store_true', help='save cropped prediction boxes')
  138. parser.add_argument('--nosave', action='store_true', help='do not save images/videos')
  139. parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --class 0, or --class 0 2 3')
  140. parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS')
  141. parser.add_argument('--augment', action='store_true', help='augmented inference')
  142. parser.add_argument('--update', action='store_true', help='update all models')
  143. parser.add_argument('--project', default='runs/detect', help='save results to project/name')
  144. parser.add_argument('--name', default='exp', help='save results to project/name')
  145. parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
  146. parser.add_argument('--line-thickness', default=3, type=int, help='bounding box thickness (pixels)')
  147. parser.add_argument('--hide-labels', default=False, action='store_true', help='hide labels')
  148. parser.add_argument('--hide-conf', default=False, action='store_true', help='hide confidences')
  149. opt = parser.parse_args()
  150. print(opt)
  151. check_requirements(exclude=('tensorboard', 'pycocotools', 'thop'))
  152. with torch.no_grad():
  153. if opt.update: # update all models (to fix SourceChangeWarning)
  154. for opt.weights in ['yolov5s.pt', 'yolov5m.pt', 'yolov5l.pt', 'yolov5x.pt']:
  155. detect(opt=opt)
  156. strip_optimizer(opt.weights)
  157. else:
  158. detect(opt=opt)