Beta
Explore
Marketplace
Neural Labs
Chat
Wallet
Docs
HALLUCINOGEN – Dataset by MM-Hallu | AlphaNeural AI
You can deploy this model and start earning money today!
MM-Hallu
/
HALLUCINOGEN
like
0
visual-question-answering
en
mit
10K<n<100K
parquet
image
text
datasets
dask
polars
mlcroissant
2412.20622
us
hallucination-evaluation
object-hallucination
POPE
multimodal
vision-language-model
COCO
Views
No views yet
Model card
Files and Versions
Community
API
HALLUCINOGEN
Comprehensive benchmark for evaluating object hallucination in LVLMs. Extends POPE-style questions across 4 task types and 3 POPE splits, producing 36,000 image-question pairs from COCO images.
Fields
Field Description
image COCO image
question_id Unique question ID
coco_image_id COCO image ID
question Base POPE question
label Ground truth: "yes" or "no"
task Task type: identification, localization, visual_contextual, counterfactual… See the full description on the dataset page:
https://huggingface.co/datasets/MM-Hallu/HALLUCINOGEN
.
