Inference with a low quality low resolution video, got a bad quality output.

[chouti]

I have a low-quality, low-resolution video, such as a CCTV camera recording in a dark and blurry environment.

After processing it with a pre-trained model, the quality of the output is still unsatisfactory. In the results, when there is a light source in the content, it appears overexposed as a purple circle. Additionally, other objects in the video remain unclear.

I have tried adjusting the max_seq_len parameter from 1 to 15, but this doesn't seem to have improved the situation.