local-vision-models/lama-1024-coreml

Big LaMa, Core ML, 1024 px

Big LaMa (Suvorov et al., "Resolution-robust Large Mask Inpainting with Fourier Convolutions", WACV 2022) converted to Core ML at 1024 x 1024.

• Code: https://github.com/advimman/lama (Apache-2.0)
• Weights: https://huggingface.co/smartywu/big-lama (Apache-2.0)
• Inputs: image (RGB, 1024 x 1024) and mask (grayscale, 255 marks the area to fill)
• Output: output (RGB, the image with the area filled)