IJCAI2023-CoNR

Why we included this project

CoNR is the official implementation of a paper that tackles one of the most tedious parts of 2D animation: redrawing a character for every frame. Give it a small set of reference images, the character sheet, and it can render that character in new poses, so you can drive a character through a dance or action sequence without hand-drawing each frame. Instead of fitting a universal body model like SMPL, which works for real human shapes but not anime hairstyles and clothing, it uses a compact landmark encoding that keeps fine details like accessories. The repo includes pretrained weights and a large open character-sheet dataset, so you can reproduce the paper's results or try your own characters. It's research code aimed at artists and researchers exploring automated in-betweening and pose-driven animation, not a turnkey production tool.