Research map / 3D vision / Face and shape
Single Image 3D Interpreter Network
Jiajun Wu, Tianfan Xue, Joseph J. Lim, Yuandong Tian, Joshua B. Tenenbaum, Antonio Torralba, William T. Freeman
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 keypoint refinement network (ablation) | 3D-INN keypoint estimator · FLIC | PCK: about 2% for a normalized distance >= 0.15 | Improved |
| add fine-tuning of the whole network with the projection layer (ablation) | 3D-INN · IKEA structure estimation | RMSE-Recall: about 5% under the RMSE threshold 0.15 | Improved |
| replace optimization-based reprojection fitting with learned 3D interpreter | 3D interpreter · synthetic heatmaps with noise | 3D structure and viewpoint accuracy | Improved |
| use 3D-INN instead of Su et al. for viewpoint estimation | viewpoint estimation · IKEA azimuth recall | average recall: about 40% (relative) | Improved |
| use 3D-INN instead of Mdshift | keypoint estimation · CUB-200-2011 keypoint estimation | PCP (%): 69.1 -> 66.7 | Hurt |
| use 3D-INN instead of Mdshift | keypoint estimation · CUB-200-2011 keypoint estimation | Average Error: 1.39 -> 1.35 | Improved |
| use 3D-INN instead of Tompson et al. on swivel chairs | keypoint estimation · Keypoint-5 keypoint estimation (Swivel Chair) | PCP (%): 69.2 -> 78.5 | Improved |
| train directly end-to-end on real 2D images without staged training (ablation) | 3D-INN · training | convergence | Hurt |
| add Batch Normalization to keypoint estimator (ablation) | keypoint estimator · training | convergence | Improved |
Closest papers by meaning
- 3D Interpreter Networks for Viewer-Centered Wireframe Modeling (IJCV 2018 · similarity 0.97)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.88)
- KeypointDETR: An End-to-End 3D Keypoint Detector (ECCV 2024 · similarity 0.88)
- 2D Features Are All You Need for 3D Shape Understanding (ECCV 2026 · similarity 0.87)
- Structured3D: A Large Photo-realistic Dataset for Structured 3D Modeling (ECCV 2020 · similarity 0.87)
- Unsupervised 3D Structure Inference from Category-Specific Image Collections (CVPR 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.
