Bai360/Cotton2
0
1# YOLOv5 🚀 by Ultralytics, GPL-3.0 license2"""3Run YOLOv5 classification inference on images, videos, directories, globs, YouTube, webcam, streams, etc.4 5Usage - sources:6 $ python classify/predict.py --weights yolov5s-cls.pt --source 0 # webcam7 img.jpg # image8 vid.mp4 # video9 screen # screenshot10 path/ # directory11 list.txt # list of images12 list.streams # list of streams13 'path/*.jpg' # glob14 'https://youtu.be/Zgi9g1ksQHc' # YouTube15 'rtsp://example.com/media.mp4' # RTSP, RTMP, HTTP stream16 17Usage - formats:18 $ python classify/predict.py --weights yolov5s-cls.pt # PyTorch19 yolov5s-cls.torchscript # TorchScript20 yolov5s-cls.onnx # ONNX Runtime or OpenCV DNN with --dnn21 yolov5s-cls_openvino_model # OpenVINO22 yolov5s-cls.engine # TensorRT23 yolov5s-cls.mlmodel # CoreML (macOS-only)24 yolov5s-cls_saved_model # TensorFlow SavedModel25 yolov5s-cls.pb # TensorFlow GraphDef26 yolov5s-cls.tflite # TensorFlow Lite27 yolov5s-cls_edgetpu.tflite # TensorFlow Edge TPU28 yolov5s-cls_paddle_model # PaddlePaddle29"""30 31import argparse32import os33import platform34import sys35from pathlib import Path36 37import torch38import torch.nn.functional as F39 40FILE = Path(__file__).resolve()41ROOT = FILE.parents[1] # YOLOv5 root directory42if str(ROOT) not in sys.path:43 sys.path.append(str(ROOT)) # add ROOT to PATH44ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative45 46from models.common import DetectMultiBackend47from utils.augmentations import classify_transforms48from utils.dataloaders import IMG_FORMATS, VID_FORMATS, LoadImages, LoadScreenshots, LoadStreams49from utils.general import (LOGGER, Profile, check_file, check_img_size, check_imshow, check_requirements, colorstr, cv2,50 increment_path, print_args, strip_optimizer)51from utils.plots import Annotator52from utils.torch_utils import select_device, smart_inference_mode53 54 55@smart_inference_mode()56def run(57 weights=ROOT / 'yolov5s-cls.pt', # model.pt path(s)58 source=ROOT / 'data/images', # file/dir/URL/glob/screen/0(webcam)59 data=ROOT / 'data/coco128.yaml', # dataset.yaml path60 imgsz=(224, 224), # inference size (height, width)61 device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu62 view_img=False, # show results63 save_txt=False, # save results to *.txt64 nosave=False, # do not save images/videos65 augment=False, # augmented inference66 visualize=False, # visualize features67 update=False, # update all models68 project=ROOT / 'runs/predict-cls', # save results to project/name69 name='exp', # save results to project/name70 exist_ok=False, # existing project/name ok, do not increment71 half=False, # use FP16 half-precision inference72 dnn=False, # use OpenCV DNN for ONNX inference73 vid_stride=1, # video frame-rate stride74):75 source = str(source)76 save_img = not nosave and not source.endswith('.txt') # save inference images77 is_file = Path(source).suffix[1:] in (IMG_FORMATS + VID_FORMATS)78 is_url = source.lower().startswith(('rtsp://', 'rtmp://', 'http://', 'https://'))79 webcam = source.isnumeric() or source.endswith('.streams') or (is_url and not is_file)80 screenshot = source.lower().startswith('screen')81 if is_url and is_file:82 source = check_file(source) # download83 84 # Directories85 save_dir = increment_path(Path(project) / name, exist_ok=exist_ok) # increment run86 (save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir87 88 # Load model89 device = select_device(device)90 model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data, fp16=half)91 stride, names, pt = model.stride, model.names, model.pt92 imgsz = check_img_size(imgsz, s=stride) # check image size93 94 # Dataloader95 bs = 1 # batch_size96 if webcam:97 view_img = check_imshow(warn=True)98 dataset = LoadStreams(source, img_size=imgsz, transforms=classify_transforms(imgsz[0]), vid_stride=vid_stride)99 bs = len(dataset)100 elif screenshot:101 dataset = LoadScreenshots(source, img_size=imgsz, stride=stride, auto=pt)102 else:103 dataset = LoadImages(source, img_size=imgsz, transforms=classify_transforms(imgsz[0]), vid_stride=vid_stride)104 vid_path, vid_writer = [None] * bs, [None] * bs105 106 # Run inference107 model.warmup(imgsz=(1 if pt else bs, 3, *imgsz)) # warmup108 seen, windows, dt = 0, [], (Profile(), Profile(), Profile())109 for path, im, im0s, vid_cap, s in dataset:110 with dt[0]:111 im = torch.Tensor(im).to(model.device)112 im = im.half() if model.fp16 else im.float() # uint8 to fp16/32113 if len(im.shape) == 3:114 im = im[None] # expand for batch dim115 116 # Inference117 with dt[1]:118 results = model(im)119 120 # Post-process121 with dt[2]:122 pred = F.softmax(results, dim=1) # probabilities123 124 # Process predictions125 for i, prob in enumerate(pred): # per image126 seen += 1127 if webcam: # batch_size >= 1128 p, im0, frame = path[i], im0s[i].copy(), dataset.count129 s += f'{i}: '130 else:131 p, im0, frame = path, im0s.copy(), getattr(dataset, 'frame', 0)132 133 p = Path(p) # to Path134 save_path = str(save_dir / p.name) # im.jpg135 txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}') # im.txt136 137 s += '%gx%g ' % im.shape[2:] # print string138 annotator = Annotator(im0, example=str(names), pil=True)139 140 # Print results141 top5i = prob.argsort(0, descending=True)[:5].tolist() # top 5 indices142 s += f"{', '.join(f'{names[j]} {prob[j]:.2f}' for j in top5i)}, "143 144 # Write results145 text = '\n'.join(f'{prob[j]:.2f} {names[j]}' for j in top5i)146 if save_img or view_img: # Add bbox to image147 annotator.text((32, 32), text, txt_color=(255, 255, 255))148 if save_txt: # Write to file149 with open(f'{txt_path}.txt', 'a') as f:150 f.write(text + '\n')151 152 # Stream results153 im0 = annotator.result()154 if view_img:155 if platform.system() == 'Linux' and p not in windows:156 windows.append(p)157 cv2.namedWindow(str(p), cv2.WINDOW_NORMAL | cv2.WINDOW_KEEPRATIO) # allow window resize (Linux)158 cv2.resizeWindow(str(p), im0.shape[1], im0.shape[0])159 cv2.imshow(str(p), im0)160 cv2.waitKey(1) # 1 millisecond161 162 # Save results (image with detections)163 if save_img:164 if dataset.mode == 'image':165 cv2.imwrite(save_path, im0)166 else: # 'video' or 'stream'167 if vid_path[i] != save_path: # new video168 vid_path[i] = save_path169 if isinstance(vid_writer[i], cv2.VideoWriter):170 vid_writer[i].release() # release previous video writer171 if vid_cap: # video172 fps = vid_cap.get(cv2.CAP_PROP_FPS)173 w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))174 h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))175 else: # stream176 fps, w, h = 30, im0.shape[1], im0.shape[0]177 save_path = str(Path(save_path).with_suffix('.mp4')) # force *.mp4 suffix on results videos178 vid_writer[i] = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h))179 vid_writer[i].write(im0)180 181 # Print time (inference-only)182 LOGGER.info(f"{s}{dt[1].dt * 1E3:.1f}ms")183 184 # Print results185 t = tuple(x.t / seen * 1E3 for x in dt) # speeds per image186 LOGGER.info(f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {(1, 3, *imgsz)}' % t)187 if save_txt or save_img:188 s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''189 LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")190 if update:191 strip_optimizer(weights[0]) # update model (to fix SourceChangeWarning)192 193 194def parse_opt():195 parser = argparse.ArgumentParser()196 parser.add_argument('--weights', nargs='+', type=str, default=ROOT / 'yolov5s-cls.pt', help='model path(s)')197 parser.add_argument('--source', type=str, default=ROOT / 'data/images', help='file/dir/URL/glob/screen/0(webcam)')198 parser.add_argument('--data', type=str, default=ROOT / 'data/coco128.yaml', help='(optional) dataset.yaml path')199 parser.add_argument('--imgsz', '--img', '--img-size', nargs='+', type=int, default=[224], help='inference size h,w')200 parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')201 parser.add_argument('--view-img', action='store_true', help='show results')202 parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')203 parser.add_argument('--nosave', action='store_true', help='do not save images/videos')204 parser.add_argument('--augment', action='store_true', help='augmented inference')205 parser.add_argument('--visualize', action='store_true', help='visualize features')206 parser.add_argument('--update', action='store_true', help='update all models')207 parser.add_argument('--project', default=ROOT / 'runs/predict-cls', help='save results to project/name')208 parser.add_argument('--name', default='exp', help='save results to project/name')209 parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')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 parser.add_argument('--vid-stride', type=int, default=1, help='video frame-rate stride')213 opt = parser.parse_args()214 opt.imgsz *= 2 if len(opt.imgsz) == 1 else 1 # expand215 print_args(vars(opt))216 return opt217 218 219def main(opt):220 check_requirements(exclude=('tensorboard', 'thop'))221 run(**vars(opt))222 223 224if __name__ == "__main__":225 opt = parse_opt()226 main(opt)227