MaskFlow: New Framework for Precise Regional Image Editing
A novel training framework named MaskFlow has been introduced for regional image editing, tackling issues related to accurate localization, consistent background maintenance, and smooth boundary transitions. This framework incorporates masks into both the probability path and flow-matching goals, ensuring generation occurs within the editable area while safeguarding the external source. To enhance seamless integration, a Soft-Poisson de-seaming module fine-tunes the predicted vector field during both training and sampling. Additionally, the creators developed a data synthesis pipeline to generate MEData, a mask-focused dataset. The research paper can be found on arXiv under ID 2608.06929.
Key facts
- MaskFlow is a training framework for regional image editing.
- It focuses on precise localization, consistent background preservation, and seamless boundary transitions.
- The mask is incorporated into the probability path and flow-matching objective.
- A Soft-Poisson de-seaming module refines the predicted vector field.
- A data synthesis pipeline constructs MEData, a mask-based dataset.
- The paper is available on arXiv with ID 2608.06929.
- The announcement type is cross.
- The paper is from arXiv.
Entities
Institutions
- arXiv