Research map / 3D vision / Human pose estimation
COIN: Control-Inpainting Diffusion Prior for Human and Camera Motion Estimation
Jiefeng Li, Ye Yuan, Davis Rempe, Haotian Zhang, Pavlo Molchanov, Cewu Lu, Jan Kautz, Umar Iqbal
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the controlled denoiser (controlled sampling) to COIN-SDS (ablation) | COIN · RICH | W-MPJPE (mm): 825.0 -> 254.5 | Improved |
| add soft inpainting (ablation) | COIN · RICH | PA-MPJPE (mm): 47.6 -> 42.9 | Improved |
| add soft inpainting (ablation) | COIN · RICH | W-MPJPE (mm): 325.8 -> 254.5 | Improved |
| use dynamic control (previous optimized motion as control signal) (ablation) | COIN · RICH | W-MPJPE (mm): 293.8 -> 254.5 | Improved |
| add human-scene relation loss (ablation) | COIN · RICH | W-MPJPE (mm): 273.0 -> 254.5 | Improved |
| add human-scene relation loss (ablation) | COIN · HCM camera motion | ATE-S (mm): 402.1 -> 385.9 | Improved |
| use COIN-SDS instead of vanilla SDS (ablation) | COIN · RICH | W-MPJPE (mm): 1453.5 -> 254.5 | Improved |
| use COIN-SDS instead of noise optimization of the diffusion latent (ablation) | COIN · RICH | W-MPJPE (mm): 414.8 -> 254.5 | Improved |
| add SDS loss term to the objective (ablation) | COIN · RICH | W-MPJPE (mm): 480.6 -> 254.5 | Improved |
| add smoothness loss (ablation) | COIN · RICH | W-MPJPE (mm): 256.1 -> 254.5 | No clear effect |
| use guided sampling for the diffusion prior (ablation) | COIN · RICH | W-MPJPE (mm): 254.5 -> 1384.6 | Hurt |
| replace DROID-SLAM with ParticleSfM for camera init (ablation) | COIN · EMDB (converged subset) | W-MPJPE (mm): 350.1 -> 330.9 | Improved |
Closest papers by meaning
- Move-in-2D: 2D-Conditioned Human Motion Generation (CVPR 2025 · similarity 0.86)
- TokenMotion: Decoupled Motion Control via Token Disentanglement for Human-centric Video Generation (CVPR 2025 · similarity 0.86)
- HumanVid: Demystifying Training Data for Camera-controllable Human Image Animation (NeurIPS 2024 · similarity 0.86)
- AC3D: Analyzing and Improving 3D Camera Control in Video Diffusion Transformers (CVPR 2025 · similarity 0.86)
- Human Motion Diffusion as a Generative Prior (ICLR 2024 · similarity 0.86)
- MotionPro: A Precise Motion Controller for Image-to-Video Generation (CVPR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
