Localization-guided image editing via cross-attention refinement

Paper Title:

LocInv: Localization-aware Inversion for Text-Guided Image Editing

Published on:

2 May 2024

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Chuanming Tang,

Kai Wang,

Fei Yang,

Joost van de Weijer

Bullets

Key Details

•

Uses localization priors like segmentation maps to guide cross-attention

•

Refines attention maps during diffusion model denoising phases

•

Enables precise editing of specific objects in image

•

Prevents unintended changes to unrelated regions

•

Evaluated on COCO images, shows quantitative and qualitative improvements

Explore the topics in this paper

attention

image-editing

localization

segmentation

text2image

AI generated summary

Localization-guided image editing via cross-attention refinement

This paper proposes a technique called Localization-aware Inversion (LocInv) that uses segmentation maps or bounding boxes to refine cross-attention maps in text-to-image models. This allows for more precise, fine-grained image editing focused on particular objects, while preventing unintended changes to other regions.