ankitmahala07/sam3d-objects-mac

This project is an unofficial port of Meta's SAM 3D Objects model that lets it run on Apple Silicon Macs instead of requiring an NVIDIA graphics card. You give it a photo of an object, and it produces a 3D representation of that object called a gaussian splat, along with a textured mesh file in the common .glb format. The original model only works with NVIDIA's CUDA technology, so this fork rewrites the parts that depended on it to work with Apple's own graphics processing instead. To use it you need an Apple Silicon Mac such as an M1, M2, M3, or M4 chip, with around 24 to 30 gigabytes of memory recommended, plus about 12 gigabytes of free disk space for the model's weights. There is a guided setup script for people who are not comfortable with the command line, which walks through installing everything and downloading the required files, as well as a manual install path for those who prefer to run the steps themselves. The model weights themselves are not included in this repository and must be downloaded separately from Hugging Face after requesting access, since they belong to Meta and are governed by Meta's own license. Once set up, running the tool is a single command that asks for your source image or images, where to save the output, a quality level that trades speed for detail, and which type of 3D mesh export you want. It supports feeding in a single photo, multiple photos of the same object from different angles for better accuracy, or a whole folder of images to process one after another automatically. Background removal from the photo happens automatically, so there is no need to manually cut out the object first. The tool produces several output files, including the extracted object image, the raw 3D splat data, and one or two versions of the final 3D mesh depending on whether you want a game-ready optimized version, an unoptimized high-detail version, or both. It also lets you regenerate just the mesh from previously saved data without repeating the more expensive initial 3D generation step.