fudan-generative-vision/champ
A diffusion-based human image animation system that generates coherent motion videos from a reference image using 3D parametric model guidance.
Velocity · 7d
+5.2
★ / day
Trend
→steady
star history
Champ is a video generation model that animates a single human image by leveraging 3D parametric representations like SMPL to guide motion generation. The method ensures temporal consistency and user-controllable animation through parametric guidance, allowing precise control over generated human motion sequences. Released at ECCV 2024 with training code and sample data.