IABench

# IABench: Incremental Attribution Benchmark Official dataset repository for the CVPR 2026 paper: **"IncreFA: Breaking the Static Wall of Generative Model Attribution"** [Project Page](https://github.com/Ant0ny44/IncreFA) | [Paper Link](https://github.com/Ant0ny44/IncreFA) | [Code](https://github.com/Ant0ny44/IncreFA) ## 1. Dataset Description As AI generative models evolve at an unprecedented speed, image attribution has become a moving target. New diffusion, adversarial, and autoregressive generators appear almost monthly. **IABench** (Incremental Attribution Benchmark) is specifically designed to redefine attribution as a **structured incremental learning problem**. ## 2. Dataset Structure ### 2.1 Data Instances Each instance contains an image, its generator label, and the original filename. ### 2.2 Data Fields - `image`: A `PIL.Image.Image` object representing the generated image. - `label`: A `string` indicating the generative model (e.g., 'SDXL', 'Flux.1'). - `file_name`: A `string` of the original image path in the repository. ### 2.3 Data Splits The dataset follows a **temporally ordered open-set protocol** as described in the paper. By default, it is provided as a `train` split, which can be further partitioned for incremental learning tasks. ## 3. Usage ```python from modelscope.hub.snapshot_download import snapshot_download # Target local directory local_dir = "/your/local/path/IABench" snapshot_download( repo_id='Ant0ny/IABench', local_dir=local_dir, repo_type='dataset' ) import os from datasets import load_dataset # Path to the downloaded 'data' folder data_dir = "/your/local/path/IABench/data" # Collect shards data_files = sorted([ os.path.join(data_dir, f) for f in os.listdir(data_dir) if f.endswith('.arrow') ]) # Load dataset dataset = load_dataset("arrow", data_files=data_files, split="train") ``` ## 4. Maintenance ### 4.1 Citation If you find this dataset helpful, please cite our work: ```bibtex @inproceedings{qin2026increfa, title={IncreFA: Breaking the Static Wall of Generative Model Attribution}, author={Haotian Qin, Dongliang Chang, Yueying Gao, Lei Chen, and Zhanyu Ma}, booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, year={2026} } ``` ### 4.2 Contact - **Author**: Haotian Qin - **Email**: [e-mail] - **GitHub**: [Ant0ny44/IncreFA](https://github.com/Ant0ny44/IncreFA) ```