zc-alexfan/arctic
ARCTIC is a dataset and benchmark for training neural networks to reconstruct 3D hand poses and hand-object interactions from multi-view RGB/D/mocap data.
Velocity · 7d
+0.4
★ / day
Trend
→steady
star history
This repository provides the official ARCTIC dataset for dexterous bimanual hand-object manipulation, along with preprocessing pipelines, visualization tools, and baseline model implementations in PyTorch. The dataset includes RGB images, depth maps, segmentation masks, and motion capture data. Researchers can use this to train hand tracking, pose estimation, and 3D reconstruction models.