noise when decoding hevc yuv420p10le videos with device="cuda"

[hanchchch]

🐛 Describe the bug

when I try to decode videos with pixel format yuv420p10le like below,

# ffprobe result
  Stream #0:0: Video: hevc (Main 10), yuv420p10le(tv, bt709/bt709/unknown), 1920x1080, SAR 1:1 DAR 16:9, 23.98 fps, 23.98 tbr, 1k tbn (default)

# decoding script
decoder = VideoDecoder("./test.mp4", device="cuda")
frames = decoder.get_frames_in_range(0, 100)

cv2.imwrite("frame.png", frames.data.permute(0, 2, 3, 1).cpu().numpy()[0])

decoded frames appears like this, with full of noise:

but with h264 yuv420p videos or if I use cpu decoding it works fine.

  Stream #0:0: Video: h264 (High), yuv420p(tv, bt709, progressive), 1920x816 [SAR 1:1 DAR 40:17], 23.98 fps, 23.98 tbr, 1k tbn (default)

why is this happening?

Versions

0.2.1

[haixuanTao]

I'm not super familiar with torch codec but it seems that the bitdepth is hardcoded as uint8 (#585) and so trying with yuv420p10le (10bit) encoding should result in either reading 2 pixels over one or truncation.