Beta
Explore
Marketplace
Neural Labs
Chat
Wallet
Docs
comp-ref – Dataset by aish-vat | AlphaNeural AI
You can deploy this model and start earning money today!
aish-vat
/
comp-ref
like
0
image-to-text
visual-question-answering
en
mit
n<1K
imagefolder
image
datasets
mlcroissant
us
safety
vlm
benchmark
multimodal
Views
No views yet
Model card
Files and Versions
Community
API
COMP-REF: Compositional Refusal Benchmark
A benchmark for measuring compositional refusal calibration in Vision-Language Models (VLMs).
Dataset Description
COMP-REF contains image-caption pairs where:
Images are individually benign Captions are individually benign
But their composition can suggest safety-relevant interpretations
Dataset Structure
images/ high/ # High severity scenarios subtle/ # Subtle scenarios
benchmark_final.csv # Main… See the full description on the dataset page:
https://huggingface.co/datasets/aish-vat/comp-ref
.
