🖼️ DiffSeg30k -- A multi-turn diffusion-editing dataset for localized AIGC detection

A dataset for segmenting diffusion-based edits — ideal for training and evaluating models that localize edited regions and identify the underlying diffusion model, as presented in the paper DiffSeg30k: A Multi-Turn Diffusion Editing Benchmark for Localized AIGC Detection.

📁 Dataset Usage

• xxxxxxxx.image.png: Edited images. Each image may have undergone 1, 2, or 3 editing operations.
• xxxxxxxx.mask.png: The corresponding mask indicating edited regions, where pixel values encode both the type of edit and the diffusion model used.

Load images and masks as follows:

from datasets import load_dataset
dataset = load_dataset("Chaos2629/Diffseg30k", split="train")
image, mask = dataset[0]['image'], dataset[0]['mask']

🧠 Mask Annotation

Each mask is a grayscale image (PNG format), where pixel values correspond to a specific editing model. The mapping is as follows:

📌 Notes

• Each edited image may be edited multiple turns, so the corresponding mask may contain several different label values ranging from 0 to 8.