EC-Compliance-50K: E-Commerce Compliance Inpainting Benchmark
收藏官方服务：
资源简介：
A benchmark dataset of 50,000 image quadruplets for evaluating automated removal of policy-violating visual elements (logos, QR codes, stickers, text, watermarks) from e-commerce product images. Each sample includes: (1) clean ground-truth image, (2) image with synthetically overlaid violation, (3) pixel-precise binary mask, and (4) coarse bounding-box mask. Designed for training and evaluating inpainting methods in the compliance-driven e-commerce setting.
提供机构：
Zenodo创建时间：
2026-07-30
