[KongchangAI]
ModelRestora-Flow求解器

Restora-Flow

An inference-time constrained solver that anchors known-region pixel values to observed measurements during the flow matching generation process. This enables an unconditional generative prior to perform mask-constrained image inpainting under zero-shot conditions without task-specific training.

Timeline (last 90 days)

Oct 8

The team introduced the Restora-Flow solver, which leverages input masks to impose constraints on the unconditional prior to achieve zero-shot inpainting.

Unverified50%

All Facts (1)

Source Articles