Research map / 3D vision / Human pose estimation
PoseDiffusion: Solving Pose Estimation via Diffusion-aided Bundle Adjustment
Jianyuan Wang, Christian Rupprecht, David Novotny
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 geometry-guided sampling (Sampson epipolar guidance) (ablation) | PoseDiffusion · CO3Dv2 (10 frames) | mAA(30): 56.0 -> 66.5 | Improved |
| use diffusion instead of direct regression (ablation) | pose estimator with same architecture · CO3Dv2 (10 frames) | mAA(30): 45.0 -> 56.0 (without GGS) | Improved |
| apply GGS to a regression model without diffusion (ablation) | PoseReg · CO3Dv2 (10 frames) | mAA(30): 45.0 -> 48.2 | Improved |
| add GGS for transfer to RealEstate10k (ablation) | PoseDiffusion · RealEstate10k (trained on CO3Dv2, 10 frames) | mAA(30): 18.7 -> 48.0 | Improved |
| add GGS on unseen categories (ablation) | PoseDiffusion · CO3Dv2 unseen categories (10 frames) | mAA(30): 40.1 -> 50.8 | Improved |
| use DINO ViT-S16 backbone instead of supervised ResNet50 (ablation) | PoseDiffusion · CO3Dv2 (10 frames) | mAA(30): 63.1 -> 66.5 | Improved |
| use self-supervised DINO ResNet50 instead of supervised ResNet50 (ablation) | PoseDiffusion · CO3Dv2 (10 frames) | mAA(30): 63.1 -> 64.3 | Improved |
| use 100 diffusion steps instead of 30 (ablation) | PoseDiffusion · CO3Dv2 (10 frames) | mAA(30): 62.5 -> 66.5 | Improved |
| increase diffusion steps from 100 to 500 (ablation) | PoseDiffusion · CO3Dv2 (10 frames) | mAA(30): 66.5 -> 65.3 | Hurt |
| keep background pixels visible to the model (ablation) | PoseDiffusion · CO3Dv2 (10 frames) | mAA(30): 57.0 -> 66.5 | Improved |
| replace predicted focal lengths with ground truth in NeRF training (ablation) | PoseDiffusion · CO3Dv2 NeRF novel view synthesis (10 frames) | PSNR: 24.37 -> 24.72 | No clear effect |
| use PoseDiffusion cameras instead of COLMAP+SPSG for NeRF with few views | NeRF on CO3Dv2 · CO3Dv2 NeRF novel view synthesis (10 frames) | PSNR: 15.78 -> 24.37 | Improved |
Later papers that compared with this method
6 later accepted papers report PoseDiffusion as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.
- Pose without Guesses: Generalizable Object Pose Estimation From a Single Reference (IJCV 2026 · better on 2 of 2 shared results)
- PosePlaner: Denoising Any Feedforward Pose Predictor from Pairwise Planar Geometry (NeurIPS 2026 · better on 1 of 1 shared result)
- High-Resolution Open-Vocabulary Object 6D Pose Estimation (TPAMI 2025 · better on 1 of 1 shared result)
- 3D Congealing: 3D-Aware Image Alignment in the Wild (ECCV 2024 · better on 3 of 3 shared results)
- VGGSfM: Visual Geometry Grounded Deep Structure From Motion (CVPR 2024 · better on 2 of 2 shared results)
- DUSt3R: Geometric 3D Vision Made Easy (CVPR 2024 · better on 1 of 1 shared result)
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
- Object Pose Estimation via the Aggregation of Diffusion Features (CVPR 2024 · similarity 0.89)
- Cameras as Rays: Pose Estimation via Ray Diffusion (ICLR 2024 · similarity 0.89)
- Confronting Ambiguity in 6D Object Pose Estimation via Score-Based Diffusion on SE(3) (CVPR 2024 · similarity 0.87)
- $\text{Di}^2\text{Pose}$: Discrete Diffusion Model for Occluded 3D Human Pose Estimation (NeurIPS 2024 · similarity 0.87)
- DiffusionSfM: Predicting Structure and Motion via Ray Origin and Endpoint Diffusion (CVPR 2025 · similarity 0.87)
- 6D-Diff: A Keypoint Diffusion Framework for 6D Object Pose Estimation (CVPR 2024 · similarity 0.87)
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.
