SignalTech/AI
LensErase: A Foundation Model for Erase Anything
LensErase: A Foundation Model for Erase Anything
Context
What happened: Researchers introduced LensErase, a mask-based "Erase Anything" foundation model pretrained on a large paired dataset and refined with reinforcement learning for removal tasks that preserve surrounding content.
Key numbers: LensErase-100M contains 101.07 million paired source-mask-target examples.
Why it matters: The authors report substantial improvements under reference-based objective metrics and reference-free VLM evaluation, and position LensErase as a foundation for erase-related applications and systems.
What changed
Researchers introduced LensErase, a mask-based "Erase Anything" foundation model pretrained on a large paired dataset and refined with reinforcement learning for removal tasks that preserve surrounding content.
Why it matters
The authors report substantial improvements under reference-based objective metrics and reference-free VLM evaluation, and position LensErase as a foundation for erase-related applications and systems.