我有下面的代码：

# Setup Video capture
cap = cv2.VideoCapture(args['video_path'])
# Read a frame from the video
ret, frame = cap.read()
fps = cap.get(cv2.CAP_PROP_FPS)

'''
Print some information about the video
'''
print('\n#######################################')
print(f'Video Resolution: {frame.shape}')
print(f'Video FPS: {fps:.2f}')
total_num_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
print(f'Total video frames: {total_num_frames}')
print('#######################################\n')

input('Press Enter to start processing video...\n')

'''
Video Writer
    - output path
    - codec
    - FPS
    - resolution (width,height)    
'''
width = frame.shape[1] #output width
height = frame.shape[0] #output height

fourcc = cv2.VideoWriter_fourcc(*'XVID') # try *'mp4v' instead of XVID
out = cv2.VideoWriter('./demo_output/demo_result.avi', fourcc, 30, (width, height))

# Set current_frame progress to 0 before main loop
current_num_frame = 0
while True:
    # Loop through each frame of the video capture
    try:
        if width < 1920 and height < 1080:
            print('Resolution too small')
            break
        elif width >=3840 and height >=2160:
            ret, frame = cap.read()
            scale_factor = 0.5
            frame = cv2.resize(frame, (0, 0), fx=scale_factor, fy=scale_factor) # interpolation=cv2.INTER_LANCZOS4
            ori_img = frame.copy()
            print('4K Res check: True')
        else:
            ret, frame = cap.read()
            print('Other res chech: True')
    except:
        print("Finished")
        cap.release()
        break

    frame = frame.copy()
    image = tensor_transform(frame)
    image = img_transform(image).unsqueeze(0)
    print('Img transformations: True')

    with torch.no_grad():
        d6 = model(image)
        print('Model output: True')

        count, pred_kpoint = counting(d6) # (540, 960), count = int(scalar)
        # point_map = generate_point_map(pred_kpoint) # (540, 960, 3)
        box_img = generate_bounding_boxes(pred_kpoint, frame) # (540, 960, 3)
        print('box_img output: True')
        # show_fidt = show_fidt_func(d6.data.cpu().numpy()) # (540, 960, 3)

        # res1 = np.hstack((ori_img, show_fidt)) # (540, 1920, 3)
        # res2 = np.hstack((box_img, point_map)) # (540, 1920, 3)
        # res = np.vstack((res1, res2)) # (1080, 1920, 3)

        cv2.putText(box_img, "Count:" + str(count), (30, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
        # cv2.imwrite('./demo_output/demo.jpg', res)
        '''write in out_video'''
        out.write(box_img)
        print('Video write: True')
    current_num_frame += 1
    print(f"Pred: {int(count)}, frame: {current_num_frame} / {total_num_frames}")

而我的目标是读取视频的宽度和高度（分辨率）。根据分辨率，我不想继续，或者缩小视频，或者什么都不做，继续。
在4K视频上，它不起作用，但它确实进入了if语句，因为它是真的。print语句表明。
1080p视频在另一方面工作。它不会被缩小，VideoWriter会产生一个工作。avi文件。
如果我删除if语句，只保留原始视频中的缩小步骤，它仍然不起作用。

while True:
    # Loop through each frame of the video capture
    try:
        ret, frame = cap.read()
        scale_factor = 0.5
        frame = cv2.resize(frame, (0, 0), fx=scale_factor, fy=scale_factor) # interpolation=cv2.INTER_LANCZOS4
        # ori_img = frame.copy()
        print('4K Res check: True')
    except:
        print("Finished")
        cap.release()
        break

    frame = frame.copy()
    image = tensor_transform(frame)
    image = img_transform(image).unsqueeze(0)

在我搞砸之前的原始代码只是缩小视频，它曾经工作过。但是当我用1080p而不是4K来尝试它时，降尺度正在破坏质量。所以我想把一个分辨率检查，以阻止它从降频，如果视频是1080p。
最后一件事是，代码不会产生错误，一切似乎都完美地工作：

编辑：

with torch.no_grad():
    d6 = model(image)

    count, pred_kpoint = counting(d6) # (540, 960), count = int(scalar)
    point_map = generate_point_map(pred_kpoint) # (540, 960, 3)
    box_img = generate_bounding_boxes(pred_kpoint, frame) # (540, 960, 3)
    show_fidt = show_fidt_func(d6.data.cpu().numpy()) # (540, 960, 3)

    res1 = np.hstack((ori_img, show_fidt)) # (540, 1920, 3)
    res2 = np.hstack((box_img, point_map)) # (540, 1920, 3)
    res = np.vstack((res1, res2)) # (1080, 1920, 3)

    cv2.putText(res, "Count:" + str(count), (30, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
    # cv2.imwrite('./demo_output/demo.jpg', res)
    '''write in out_video'''
    out.write(res)

如果我写res变量，它是输出的组合（这是以前的做法），那么它在4K和1080P上都能工作。但是对于我的例子，我只需要box_img变量。所以当我把它放进去。write（），它将停止工作。
我猜这与高度和宽度的形状有关。