Research map / 3D vision / Face and shape
Neural Non-Rigid Tracking
Aljaz Bozic, Pablo Palafox, Michael Zollhöfer, Angela Dai, Justus Thies, Matthias Niessner
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 graph and warp losses backpropagated through differentiable solver (ablation) | Correspondence network · DeepDeform tracking | EPE 3D (mm): 44.05 -> 36.96 | Improved |
| Add self-supervised correspondence weighting (ablation) | Phi_c+g+w · DeepDeform tracking | Graph Error 3D (mm): 54.24 -> 31.00 | Improved |
| Use self-supervised instead of supervised (BCE) weighting (ablation) | Phi_c+g+w · DeepDeform tracking | EPE 3D (mm): 27.42 -> 26.29 | Improved |
| Add supervised weighting to correspondence-loss-only model (ablation) | Phi_c · DeepDeform tracking | EPE 3D (mm): 44.05 -> 28.95 | Improved |
| Use full end-to-end method instead of Phi_c only (ablation) | Reconstruction pipeline · DeepDeform reconstruction | deformation error (mm): 54.85 -> 28.72 | Improved |
| Use bilinear instead of nearest-neighbor depth sampling (ablation) | Solver · DeepDeform tracking | EPE 3D: 5.8% lower EPE 3D | Improved |
| Use ARAP edge re-weighting (ablation) | Solver · DeepDeform tracking | EPE 3D: 0.8% lower EPE 3D | No clear effect |
| Use coarser deformation graph (15 cm instead of 5 cm node coverage) (ablation) | Solver · DeepDeform tracking | Graph Error 3D: 28.46% higher Graph Error 3D | Hurt |
| Unroll 3 Gauss-Newton iterations instead of 1 (ablation) | Solver · DeepDeform tracking | EPE 3D: 24.9% lower EPE 3D | Improved |
| Use 4 Gauss-Newton iterations instead of 3 (ablation) | Solver · DeepDeform tracking | EPE 3D: 0.04% lower EPE 3D | No clear effect |
| Denser keyframes (every 50 instead of 100 frames) (ablation) | Reconstruction pipeline · DeepDeform reconstruction | deformation error (mm): 30.70 -> 28.72 | Improved |
| Replace DeepDeform sparse correspondence network with dense PWC-Net based prediction | Non-rigid reconstruction · DeepDeform reconstruction | runtime per keyframe (ms): 2299 -> 27 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DeepDeform — DeepDeform: Learning Non-Rigid RGB-D Reconstruction With Semi-Supervised Data (CVPR 2020) · better on 2 of 2 shared results
- VolumeDeform — VolumeDeform: Real-Time Volumetric Non-rigid Reconstruction (ECCV 2016) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Neural Deformation Graphs for Globally-Consistent Non-Rigid Reconstruction (CVPR 2021 · similarity 0.90)
- Learning to Optimize Non-Rigid Tracking (CVPR 2020 · similarity 0.90)
- DeepDeform: Learning Non-Rigid RGB-D Reconstruction With Semi-Supervised Data (CVPR 2020 · similarity 0.88)
- Large-Displacement 3D Object Tracking with Hybrid Non-local Optimization (ECCV 2022 · similarity 0.86)
- STaR: Self-Supervised Tracking and Reconstruction of Rigid Objects in Motion With Neural Rendering (CVPR 2021 · similarity 0.86)
- Motion2VecSets: 4D Latent Vector Set Diffusion for Non-rigid Shape Reconstruction and Tracking (CVPR 2024 · 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.
