Research map / 3D vision / 3D generation
Unsupervised Monocular 3D Keypoint Discovery from Multi-View Diffusion Priors
Subin Jeon, In Cho, Junyoung Hong, Woong Oh Cho, Seon Joo Kim
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 |
|---|---|---|---|
| use multi-view diffusion features instead of DINOv2 features (ablation) | KeyDiff3D · Human3.6M | MPJPE: 136.17 -> 121.34 | Improved |
| use multi-view diffusion features instead of ResNet50 features (ablation) | KeyDiff3D · Human3.6M | MPJPE: 138.97 -> 121.34 | Improved |
| use multi-view diffusion features instead of CLIP features (ablation) | KeyDiff3D · Human3.6M | MPJPE: 143.23 -> 121.34 | Improved |
| lift via 3D feature volume instead of 2D keypoints plus triangulation (ablation) | KeyDiff3D · Human3.6M | MPJPE: 129.63 -> 121.34 | Improved |
| add generated virtual views beyond the input view (1 to 4) (ablation) | KeyDiff3D · Human3.6M | MPJPE: 166.29 -> 121.34 | Improved |
| increase virtual views from 4 to 5 (ablation) | KeyDiff3D · Human3.6M | MPJPE: 121.34 -> 119.77 | No clear effect |
| increase keypoints from 16 to 48 (ablation) | KeyDiff3D · Human3.6M | MPJPE: 127.99 -> 116.45 | Improved |
| use diffusion timestep 500 instead of 900 (ablation) | KeyDiff3D · Human3.6M | MPJPE: 131.26 -> 121.34 | Improved |
| increase training images from 2,200 to 8,800 (ablation) | KeyDiff3D · Human3.6M | MPJPE: 134.26 -> 121.34 | Improved |
| add random affine transformation to the input of reconstruction network (ablation) | KeyDiff3D · Human3.6M | MPJPE: 134.54 -> 121.34 | Improved |
| add camera rotation noise of 1.0 degree (ablation) | KeyDiff3D · Human3.6M | MPJPE: 121.34 -> 133.88 | Hurt |
| add camera rotation noise of 0.2 degree (ablation) | KeyDiff3D · Human3.6M | MPJPE: 121.34 -> 121.30 | No clear effect |
Closest papers by meaning
- MonoDiff: Monocular 3D Object Detection and Pose Estimation with Diffusion Models (CVPR 2024 · similarity 0.88)
- Unsupervised 3D Structure Inference from Category-Specific Image Collections (CVPR 2024 · similarity 0.88)
- Diffusion 3D Features (Diff3F): Decorating Untextured Shapes with Distilled Semantic Features (CVPR 2024 · similarity 0.87)
- Sculpt3D: Multi-View Consistent Text-to-3D Generation with Sparse 3D Prior (CVPR 2024 · similarity 0.87)
- Era3D: High-Resolution Multiview Diffusion using Efficient Row-wise Attention (NeurIPS 2024 · similarity 0.87)
- Sparse3D: Distilling Multiview-Consistent Diffusion for Object Reconstruction from Sparse Views (AAAI 2024 · 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.
