update

model_inference.py

@@ -96,12 +96,12 @@ model_embedding.load_state_dict(torch.load('./embedding-output/model_embedding.p
 model_embedding.to(device)
 model_embedding.eval()
 
-with open('/model_classifier.obj','rb') as file:
+with open('./model_classifier.obj','rb') as file:
   model_classifier = pickle.load(file)
 
 classes = model_classifier.__getstate__()['classes_']
 
-video = Path('/content/test_videos_2022/2022-NLS-5-NLS_05_2022_Heli_UHD_01-000140-000155-Karussell.mp4')
+video = Path('./test_videos_2022/2022-NLS-5-NLS_05_2022_Heli_UHD_01-000140-000155-Karussell.mp4')
 
 reader = cv2.VideoCapture(str(video))
 fps = reader.get(cv2.CAP_PROP_FPS)
@@ -109,18 +109,19 @@ w = int(reader.get(cv2.CAP_PROP_FRAME_WIDTH))
 h = int(reader.get(cv2.CAP_PROP_FRAME_HEIGHT))
 reader.release()
 
-imgsz = check_img_size((w, h), s=model.stride)
-dataset = LoadImages(video, img_size=imgsz, stride=model.stride, auto=model.pt)
-
 weights_path = Path('./yolov5/best.pt')
 model = DetectMultiBackend(weights_path, device=torch.device(device))
 
+imgsz = check_img_size((w, h), s=model.stride)
+dataset = LoadImages(video, img_size=imgsz, stride=model.stride, auto=model.pt)
+
 save_dir = Path('./detection-output/')
 os.makedirs(save_dir)
 
 writer = cv2.VideoWriter(str(save_dir / 'res.mp4'), cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h))
 
 for frame_n, (path, im, im0s, vid_cap, s) in enumerate(dataset):
+  print(frame_n)
   im = torch.from_numpy(im).to(device)
   im = im.half() if model.fp16 else im.float()  # uint8 to fp16/32
   im /= 255  # 0 - 255 to 0.0 - 1.0

requirements.txt

@@ -1,4 +1,5 @@
 gdown
+seaborn
 torch
 torchvision
 sklearn