Datasets:
|
Download sources/HalluSegBench.md from SteveWCG/Attn_exp_data: direct link, hf CLI and curl.
- Browser
- Download file 2.54 kB
-
https://huggingface.co/datasets/SteveWCG/Attn_exp_data/resolve/main/sources/HalluSegBench.md
- Command line
-
hf download hf://datasets/SteveWCG/Attn_exp_data/sources/HalluSegBench.md
-
curl -L -o HalluSegBench.md https://huggingface.co/datasets/SteveWCG/Attn_exp_data/resolve/main/sources/HalluSegBench.md
2.54 kB
| configs: | |
| - config_name: default | |
| data_files: | |
| - split: train | |
| path: viewer/train.parquet | |
| - split: test | |
| path: viewer/test.parquet | |
| # HalluSegBench | |
| HalluSegBench is a counterfactual segmentation reasoning dataset for evaluating pixel-grounding hallucination in vision-language segmentation models. | |
| Each example contains a factual image, a factual segmentation mask, a counterfactual image, and a counterfactual segmentation mask. The counterfactual image replaces the original target object or region with a plausible alternative while preserving the surrounding scene context. The dataset is designed to test whether a model segments objects based on visual evidence rather than language priors. | |
| ## Dataset Structure | |
| The full dataset is provided as: | |
| ```text | |
| train.zip | |
| test.zip | |
| ``` | |
| After extraction, the dataset contains: | |
| ```text | |
| train/ | |
| refer_seg/ | |
| reason_seg_v1/ | |
| reason_seg_v2/ | |
| test/ | |
| refer_seg/ | |
| reason_seg_val/ | |
| reason_seg_test/ | |
| ``` | |
| Each subset contains: | |
| ```text | |
| data.json | |
| factual_images/ | |
| factual_masks/ | |
| counterfactual_images/ | |
| counterfactual_masks/ | |
| ``` | |
| ## Data Format | |
| Each `data.json` entry includes paths and metadata for a factual-counterfactual pair, such as: | |
| ```json | |
| { | |
| "factual_image_path": "factual_images/example.jpg", | |
| "factual_mask_path": "factual_masks/example_mask.png", | |
| "counterfactual_image_path": "counterfactual_images/example_edited.png", | |
| "counterfactual_mask_path": "counterfactual_masks/example_newmask.png", | |
| "label": "target object", | |
| "ann_id": "annotation_id", | |
| "gpt_instruction": "Change target object to replacement object", | |
| "factual_label": "original target description", | |
| "counterfactual_label": "replacement target description" | |
| } | |
| ``` | |
| Some reasoning segmentation examples also include `question` and `counterfactual_question`. | |
| ## Hugging Face Viewer | |
| The files under `viewer/` are small Parquet preview subsets for the Hugging Face Dataset Viewer: | |
| ```text | |
| viewer/train.parquet | |
| viewer/test.parquet | |
| ``` | |
| **Note:** These Parquet files are only for previewing examples on Hugging Face. They are not the full dataset and do not replace `train.zip` and `test.zip`. | |
| ## Citation | |
| ```bibtex | |
| @article{li2025counterfactual, | |
| title={Counterfactual Segmentation Reasoning: Diagnosing and Mitigating Pixel-Grounding Hallucination}, | |
| author={Li, Xinzhuo and Juvekar, Adheesh and Zhang, Jiaxun and Liu, Xingyou and Wahed, Muntasir and Nguyen, Kiet A. and Shen, Yifan and Yu, Tianjiao and Lourentzou, Ismini}, | |
| journal={arXiv preprint arXiv:2506.21546}, | |
| year={2025} | |
| } | |
| ``` | |