Very large GPU memory usage (191GB OOM)

[Ney1Peter]

When running inference I encounter torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 191.91 GiB, so I would like to ask whether ReCamMaster only supports a fixed resolution (832×480) and fixed video length (81 frames), whether such a large GPU memory requirement is expected, or if I might be using the inference script incorrectly.

[scpoppin1207]

I met the same issue when I inference the demo. Did you solve this problem?