Research map / 3D vision / Face and shape
SDFit: 3D Object Pose and Shape by Fitting a Morphable SDF to a Single Image
Dimitrije Antić, Georgios Paschalidis, Shashank Tripathi, Theo Gevers, Sai Kumar Dwivedi, Dimitrios Tzionas
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 |
|---|---|---|---|
| fit a morphable SDF with render-and-compare instead of feed-forward regression | SDFit · Pix3D with 40% occlusion | Chamfer distance: 6.40 -> 3.74 (ZeroShape vs SDFit) | Improved |
| inpaint occluders before running baselines | ZeroShape · Pix3D with 40% occlusion | Chamfer distance: 6.40 -> 5.23 | Improved |
| use SDFit instead of feed-forward methods on unoccluded Pix3D | Pix3D unoccluded · Pix3D (Chair, Sofa mean) | Chamfer distance: 3.41 (TripoSR) vs 3.44 (ZeroShape) vs 3.53 (SDFit) | No clear effect |
| refine both shape and pose instead of fixing shape and optimizing pose only | SDFit · image alignment (Pascal3D+, Pix3D mean) | 2D IoU (%): 78.1 -> 85.4 (ZeroShape+RnC vs SDFit feat@S_init) | Improved |
| decorate the initial shape S_init rather than the category template T for features (ablation) | SDFit · image alignment (Pascal3D+, Pix3D mean) | 2D IoU (%): 78.8 -> 85.4 | Improved |
| reduce decoration views from 100 to 8 and diffusion steps from 100 to 10 (ablation) | SDFit shape decoration · Pix3D (Chair, Sofa) | Chamfer distance: 3.53 -> 3.58; runtime 600 -> 10 sec | No clear effect |
| use ground-truth pose instead of estimated pose (ablation) | SDFit · Pix3D image alignment | 2D IoU (%): 84.3 -> 85.6 | Improved |
| use ground-truth shape instead of retrieved shape (ablation) | SDFit · Pix3D image alignment | 2D IoU (%): 84.3 -> 79.4 | Hurt |
| increase occlusion from 0% to 60% of bounding box (ablation) | ZeroShape · Pix3D | Chamfer distance: 3.44 -> 7.45 | Hurt |
| increase occlusion from 0% to 60% of bounding box (ablation) | SDFit · Pix3D | Chamfer distance: 3.53 -> 3.83 | No clear effect |
| use SDFit instead of ZeroShape for semantic plausibility | Pix3D · Pix3D (Chair, Sofa mean) | CLIP similarity x100: 29.33 -> 30.53 | 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.
- ZeroShape — ZeroShape: Regression-based Zero-shot Shape Reconstruction (CVPR 2024) · better on 3 of 4 shared results
- SDFusion — SDFusion: Multimodal 3D Shape Completion, Reconstruction, and Generation (CVPR 2023) · 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
- 3D Bounding Box Estimation Using Deep Learning and Geometry (CVPR 2017 · similarity 0.89)
- Rethinking Reprojection: Closing the Loop for Pose-Aware Shape Reconstruction From a Single Image (ICCV 2017 · similarity 0.89)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.89)
- Sparse-view Pose Estimation and Reconstruction via Analysis by Generative Synthesis (NeurIPS 2024 · similarity 0.88)
- 3D-Aware Hypothesis & Verification for Generalizable Relative Object Pose Estimation (ICLR 2024 · similarity 0.88)
- LEAP: Liberate Sparse-View 3D Modeling from Camera Poses (ICLR 2024 · similarity 0.88)
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.
