Skip to content

feat(tasks): add MMVP task with ground truth corrections #1028

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
Luodian wants to merge 2 commits into
base: main
Choose a base branch
from
Open

feat(tasks): add MMVP task with ground truth corrections #1028

Luodian wants to merge 2 commits into from
+337 −0

Conversation

@Luodian
Copy link
Contributor

@Luodian Luodian commented Jan 22, 2026 •
edited
Loading

Summary

  • Add MMVP (Multimodal Visual Patterns) benchmark task
  • Apply verified ground truth corrections for indices 99 and 279 as documented in Issues about MMVP Dataset #1018

Description

MMVP is a benchmark that tests VLMs on "CLIP-blind pairs" - images that look similar to CLIP but have clear visual differences. The dataset contains 300 samples (150 pairs) testing 9 basic visual patterns.

Features

  1. Dataset: Loads from MMVP/MMVP on HuggingFace
  2. Metrics:
    • mmvp_accuracy: Individual question accuracy
    • mmvp_pair_accuracy: Both questions in a CLIP-blind pair must be correct (stricter metric)
  3. Ground Truth Corrections: Applies verified corrections for:
    • Index 99: Elephant tusks are long, not short (corrected from B to A)
    • Index 279: Person is standing, not sitting (corrected from B to A)

Usage

python -m lmms_eval --model <model> --tasks mmvp --batch_size 1

References

@Luodian
feat(tasks): add MMVP task with ground truth corrections
8feb37c 
Add MMVP (Multimodal Visual Patterns) benchmark task that tests VLMs
on CLIP-blind pairs - images perceived as similar by CLIP but with
clear visual differences.

Key features:
- Loads dataset from MMVP/MMVP on HuggingFace
- Reports both individual accuracy and pair accuracy metrics
- Applies verified ground truth corrections for indices 99 and 279
  as documented in issue #1018

The pair accuracy metric requires models to correctly answer BOTH
questions in each CLIP-blind pair, providing a stricter evaluation
of genuine visual understanding.

Github-Issue: #1018
@Luodian
style: apply black formatting to mmvp utils
73dd2f3
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@kcz358 kcz358 kcz358 approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@Luodian @kcz358