Research map / 3D vision / Depth and stereo
Robust Geometry-Preserving Depth Estimation Using Differentiable Rendering
Chi Zhang, Wei Yin, Gang Yu, Zhibin Wang, Tao Chen, Bin Fu, Joey Tianyi Zhou, Chunhua Shen
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 multi-view rendering consistency losses to SSI training | DPT with SSI loss · NYU/ScanNet/ETH3D/KITTI (scale alignment) | AbsRel: 27.9 -> 10.5 (NYU), 21.0 -> 12.9 (KITTI) | Improved |
| add the consistency losses on top of GP2 | GP2 · ETH3D (scale alignment) | AbsRel: 14.0 -> 10.9 | Improved |
| add multi-focal-length (MFL) loss (ablation) | consistency-loss training · mean of 4 benchmarks | AbsRel: 13.3 -> 11.7 | Improved |
| add depth consistency loss (ablation) | consistency-loss training · mean of 4 benchmarks | AbsRel: 13.4 -> 11.7 | Improved |
| add image consistency loss (ablation) | consistency-loss training · mean of 4 benchmarks | AbsRel: 12.2 -> 11.7 | Improved |
| use depth consistency instead of image consistency to select FOV (ablation) | RenderFOV · KITTI point cloud | RMSE: 5.95 -> 3.04 | Improved |
| use RenderFOV depth-consistency FOV selection instead of PCM | point cloud reconstruction · KITTI point cloud | RMSE: 7.05 -> 3.04 | Improved |
| use RenderFOV with image consistency loss instead of PCM (ablation) | point cloud reconstruction · 2D3D and NYU point cloud | RMSE: 0.67 -> 0.72 (2D3D), 0.38 -> 0.43 (NYU) | Hurt |
| learn affine coefficients self-supervised with consistency losses on unlabeled domain images | SSI-trained model · KITTI | AbsRel: 21.0 -> 13.1 (vs SA); 14.0 with SSA | Improved |
| scale-only aligned geometry-preserving model vs prior SOTA with scale-and-shift alignment | Ours (SA) · ETH3D zero-shot | AbsRel: 17.1 -> 9.1 | Improved |
Closest papers by meaning
- Learning To Recover 3D Scene Shape From a Single Image (CVPR 2021 · similarity 0.92)
- Towards Accurate Reconstruction of 3D Scene Shape From A Single Monocular Image (TPAMI 2022 · similarity 0.92)
- Virtual Normal: Enforcing Geometric Constraints for Accurate and Robust Depth Prediction (TPAMI 2021 · similarity 0.91)
- Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-Shot Cross-Dataset Transfer (TPAMI 2020 · similarity 0.91)
- Learning Non-Volumetric Depth Fusion Using Successive Reprojections (CVPR 2019 · similarity 0.90)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.89)
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.
