arctic

Why we included this project

Researchers working on hand and object pose estimation often need a benchmark where the data handling doesn't get in the way. This repo pulls the ARCTIC dataset into one place, with tooling for downloading, preprocessing, splitting, and visualizing the captures, plus PyTorch dataloaders and baseline models built on MANO and SMPL-X. The rendering pipeline for RGB, depth, and segmentation masks makes it easier to inspect MoCap sequences before training. If your group studies bimanual manipulation, hand tracking, or articulated objects, this gives you both a solid comparison point and reusable scaffolding for custom models, with hooks into the official leaderboard.