aghasalim/vlm-hallucination-eval
Vision-language hallucination measured against a hand-verified adversarial set. BLIP hallucinates 4.5% but misses 29% of present objects; CLIP verification cuts hallucination 33-40% at a real cost in recall.
ORIGINAL REPOSITORY
aghasalim/vlm-hallucination-eval ↗GitHub activity
- Stars
- 6
- Forks
- 0
- Open issues and PRs
- 0
- Repository size
- ~1,108 KB
- Default branch
- main
- Created
- Aug 13, 2026
- Repository updated
- Nov 2, 2026
- Last push
- Oct 10, 2026
- Language
- Python
- License
- MIT
Topics blip · clip · hallucination-detection · machine-learning · model-evaluation · multimodal · vision-language
Repository-wide figures from GitHub, checked Nov 6, 2026. Stars are endorsements, not a Mega AI Hub quality rating. Repository size is GitHub’s estimate; a ZIP download may differ.
Resource information
- Original repository
- aghasalim/vlm-hallucination-eval
- Credit basis
- GitHub repository owner or organization; individual authorship is not asserted
- Publisher
- aghasalim
- Upstream last modified
- Nov 2, 2026
- Official website
- https://github.com/aghasalim/vlm-hallucination-eval
- Languages
- Python
- Access
- Open original GitHub repository; publisher terms apply
- Last checked
- Nov 6, 2026
- License
- MIT
- Last record update
- Nov 6, 2026