Unexpected artifacts on REDS video with official FlashVSR inference

[Folvi1]

Hi, thanks for releasing FlashVSR.

I’m observing noticeable visual artifacts when running the official infer_flashvsr_v1.1_tiny_long_video.py script on a REDS video with bicubic degradation. The input resolution is 320×180, and the output resolution is 1280×640. I attached a few example frames below.

I also observe artifacts with infer_flashvsr_v1.1_full.py, but the attached screenshots here are from the tiny long video script.

Is this expected behavior, or could something be wrong in the inference setup?