"MM-AFED"
收藏DataCite Commons2026-03-16 更新2026-05-03 收录
下载链接：
https://ieee-dataport.org/documents/mm-afed
下载链接
链接失效反馈官方服务：
资源简介：
"We construct the MM-AFED dataset, a dedicated benchmark for locality-aware multimodal fashion editing. It contains 2,308 expert-annotated quadruples, each consisting of a source image, a natural-language editing instruction, a corresponding target image, and a fine-grained edit mask. The dataset is carefully designed to support supervised learning of precise, spatially controllable attribute editing while preserving non-target regions.MM-AFED is organized into two components. The Control_image files include the source images paired with their corresponding editing instruction text, providing multimodal inputs for the editing model. The Training_image files contain the target images and the associated edit masks, which explicitly indicate the regions where modifications should occur. This structured design enables reliable supervision for mask-aware multimodal diffusion training and systematic evaluation of locality-aware editing performance."
提供机构：
IEEE DataPort
创建时间：
2026-03-16
