Overview
LaMa (Resolution-robust Large Mask Inpainting with Fourier Convolutions) is an image inpainting tool that generalizes well to high resolutions, surpassing training resolutions. It leverages Fourier convolutions to complete images, even with periodic structures. The architecture includes convolutional neural networks and Fourier transforms to process and reconstruct masked image regions. LaMa offers functionalities such as image completion, object removal, and image restoration. It supports various environment setups including Python virtualenv, Conda, and Docker. Pre-trained models are available for download to facilitate inference, and it supports both CPU and GPU-based processing. The tool allows for training and evaluation, providing metrics on image quality.
Common tasks