Research map / 3D vision / Face and shape
Learning Joint 2D-3D Representations for Depth Completion
Yun Chen, Bin Yang, Ming Liang, Raquel Urtasun
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 stride-1 2D conv branch to the fuse block (ablation) | FuseNet · KITTI val (C=32, N=12) | RMSE (mm): 840 -> 803 | Improved |
| add stride-2 2D conv branch to the fuse block (ablation) | FuseNet · KITTI val (C=32, N=12) | RMSE (mm): 826 -> 803 | Improved |
| add 3D continuous convolution branch to the fuse block (ablation) | FuseNet · KITTI val (C=32, N=12) | RMSE (mm): 817 -> 803 | Improved |
| vary number of nearest neighbors K from 3 to 15 in continuous conv (ablation) | FuseNet · KITTI val (C=32, N=9) | RMSE (mm): 813, 810, 810, 816, 812 | No clear effect |
| train with smooth l1 instead of l2 loss (ablation) | FuseNet · KITTI val (C=64, N=12) | MAE (mm): 232 -> 197 | Improved |
| train with smooth l1 instead of l2 loss (ablation) | FuseNet · KITTI val (C=64, N=12) | RMSE (mm): 790 -> 839 | Hurt |
| fine-tune l2 model with l2 + smooth l1 loss (ablation) | FuseNet · KITTI val (C=64, N=12) | RMSE (mm): 790 -> 785 | Improved |
| fine-tune l2 model with l2 + smooth l1 loss (ablation) | FuseNet · KITTI val (C=64, N=12) | MAE (mm): 232 -> 217 | Improved |
Closest papers by meaning
- Deep Depth Completion of a Single RGB-D Image (CVPR 2018 · similarity 0.92)
- RGBD Based Dimensional Decomposition Residual Network for 3D Semantic Scene Completion (CVPR 2019 · similarity 0.90)
- Deep RGB-D Canonical Correlation Analysis For Sparse Depth Completion (NeurIPS 2019 · similarity 0.89)
- CostDCNet: Cost Volume Based Depth Completion for a Single RGB-D Image (ECCV 2022 · similarity 0.89)
- DeCoTR: Enhancing Depth Completion with 2D and 3D Attentions (CVPR 2024 · similarity 0.89)
- Completion as Enhancement: A Degradation-Aware Selective Image Guided Network for Depth Completion (CVPR 2025 · 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.
