Research map / 3D vision / Depth and stereo
KeyTr: Keypoint Transporter for 3D Reconstruction of Deformable Objects in Videos
David Novotny, Ignacio Rocco, Samarth Sinha, Alexandre Carlier, Gael Kerchenbaum, Roman Shapovalov, Nikita Smetanin, Natalia Neverova, Benjamin Graham, Andrea Vedaldi
Check 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 |
|---|---|---|---|
| Model object deformation across the whole video with a low-rank keypoint basis and optical-flow consistency | KeyTr · Animals-in-motion | ℓd-scaled₁: 0.567 -> 0.135 | Improved |
| Model object deformation across the whole video with a low-rank keypoint basis and optical-flow consistency | KeyTr · Humans-in-motion – easy | ℓd₁: 1.245 -> 0.183 | Improved |
| Use KeyTr rather than DVD for metric depth reconstruction | KeyTr · Humans-in-motion – difficult | ℓd₁: 0.684 -> 0.780 | Hurt |
| Add optimal assignment and Chamfer coverage losses (ablation) | KeyTr · Animals-in-motion | ℓd-scaled₁: 8.36 -> 0.87 | Improved |
| Add the optical-flow consistency loss to assignment and Chamfer losses (ablation) | KeyTr · Animals-in-motion | ℓd-scaled₁: 0.87 -> 0.19 | Improved |
| Add the RGB reconstruction loss to assignment and Chamfer losses without the flow loss (ablation) | KeyTr · Animals-in-motion | ℓd-scaled₁: 0.87 -> 0.13 | Improved |
| Add the RGB reconstruction loss to assignment, Chamfer, and flow losses (ablation) | KeyTr · Animals-in-motion | ℓd-scaled₁: 0.19 -> 0.14 | Hurt |
| Vary the number of shape-basis vectors; the best reported setting is K = 10 (ablation) | KeyTr · Animals-in-motion | depth errors | Mixed |
Closest papers by meaning
- Stereo4D: Learning How Things Move in 3D from Internet Stereo Videos (CVPR 2025 · similarity 0.88)
- Seurat: From Moving Points to Depth (CVPR 2025 · similarity 0.87)
- 4D3R: Motion-Aware Neural Reconstruction and Rendering of Dynamic Scenes from Monocular Videos (NeurIPS 2025 · similarity 0.87)
- DGS-LRM: Real-Time Deformable 3D Gaussian Reconstruction From Monocular Videos (NeurIPS 2025 · similarity 0.87)
- Restage4D: Reanimating Deformable 3D Reconstruction from a Single Video (NeurIPS 2025 · similarity 0.86)
- MonST3R: A Simple Approach for Estimating Geometry in the Presence of Motion (ICLR 2025 · similarity 0.86)
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.
