metadata
license: apache-2.0
library_name: coreml
tags:
- coreml
- inpainting
- object-removal
LaMa (Core ML)
A mirror of the LaMa inpainting model in Core ML format (800×800 image + mask in, image out), as converted by john-rocky from advimman/lama (Samsung AI Center).
Inputs: image (800×800 BGRA), mask (800×800 grayscale, white = area to fill). Output: output (800×800 image).
All credit to the LaMa authors: Resolution-robust Large Mask Inpainting with Fourier Convolutions (paper). LaMa's code is Apache-2.0; see LICENSE.