MaterialPalette
MaterialPalette is an official implementation of the CVPR 2024 research paper titled Extraction of Materials from a Single Real-world Image. This software tool enables the extraction of Physically Based Rendering materials from a single real-world photograph. It operates through a three-stage pipeline that begins by learning texture concepts from an input image and a user-provided region mask. These concepts are then used to generate high-quality texture images. Finally, the generated textures are decomposed into full SVBRDF material maps, specifically outputting albedo, surface normal, and roughness channels. The repository provides a complete toolkit including installation scripts for setting up the Python and PyTorch environment, pretrained concept models for immediate experimentation with texture generation, and the full pipeline for concept learning, generation, and decomposition. It also includes access to the TexSD dataset and retraining capabilities for custom applications. Designed for the computer v