open-compass / open-compass/VLMEvalKit

Small performance gap between sub and nosub on Video-MME with Qwen2.5-Omni

Open
#1,226 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
4.4k
Forks
768
Avg merge
2d 27m
Merged PRs (30d)
18

Description

Hi,

I’ve been evaluating Qwen2.5-Omni on Video-MME using VLMEvalKit, and I noticed that enabling subtitles doesn’t bring much improvement compared to the no-subtitle setting. Here are my results:

  • 8 fps

    • w/o sub: overall = 0.604
    • w/ sub: overall = 0.609
  • 64 fps

    • w/o sub: overall = 0.663
    • w/ sub: overall = 0.676

The gap between “sub” and “nosub” is quite small .
According to some blog posts and papers, I expected a clearer improvement with subtitles.

So I’d like to ask:

  1. Is this behavior normal when using VLMEvalKit?
  2. Have others observed similar results when reproducing Qwen2.5-Omni on Video-MME?

Any insights or reproduction numbers would be greatly appreciated!

Thanks!

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

The issue names no files, tests, or entry points. Reproduce Qwen2.5-Omni on Video-MME at 8 and 64 fps with and without subtitles, then compare the four reported scores and determine whether the small gap is expected or indicates a problem in VLMEvalKit; done means a supported explanation or reproducible evidence.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
computer-vision, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.