ByteDance-Seed / ByteDance-Seed/SeedVR

Does SeedVR2 Refer to frames beyond the Batch Size?

Open
#46 2 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
1.4k
Forks
82
PR merge metrics
No merged PRs in 30d

Description

I have been using SeedVR2 for video restoration via ComfyUI and have achieved excellent results; I consider it to be the best video restoration tool available today!

I have set the batch size to 5 frames.

Previously, when processing long videos, the process would occasionally be interrupted, resulting in a loss of processing time. To mitigate this, I decided to split the original long video into consecutive 5-frame clips and run the restoration process on each clip individually.

However, I observed a obvious degradation in the output quality when applying the process to these short, 5-frame clips, compared to the quality obtained when processing the full, long video.

My initial assumption was that since the batch size was 5, splitting the video into 5-frame segments would produce the same result.

Given the observed quality degradation, I now suspect that SeedVR2 references frames beyond the set batch size (5 in my case) to get context when a long video is provided. Is this correct?

Thank you!

Contributor guide

No contributing guide indexed for this repository

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.