Research map / 3D vision / Depth and stereo
Sparse-to-Dense Depth Completion Revisited: Sampling Strategy and Graph Construction
Xin Xiong, Haipeng Xiong, Ke Xian, Chen Zhao, Zhiguo Cao, Xin Li
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 GNN neighborhood construction (ablation) | Ours Baseline, random sampling · NYUDv2 test | RMSE: 0.112 -> 0.106 | Improved |
| add GNN neighborhood construction (ablation) | Ours Baseline, Golden sampling · NYUDv2 test | RMSE: 0.094 -> 0.087 | Improved |
| replace random sampling with Golden quasi-random sampling (ablation) | Ours GNN · NYUDv2 test | RMSE: 0.106 -> 0.087 | Improved |
| replace random sampling with Golden quasi-random sampling (ablation) | Ours Baseline · NYUDv2 test | RMSE: 0.112 -> 0.094 | Improved |
| use Golden instead of R2 sampling (ablation) | Ours GNN · NYUDv2 test | RMSE: 0.092 -> 0.087 | Improved |
| use Halton instead of Golden sampling (ablation) | Ours GNN · NYUDv2 test | RMSE: 0.087 -> 0.093 | Hurt |
| add GNN under KITTI-to-NYUDv2 transfer, random sampling (ablation) | Ours Baseline · NYUDv2 test (transfer from KITTI) | RMSE: 0.365 -> 0.310 | Improved |
| use Golden instead of random sampling under transfer (ablation) | Ours GNN · NYUDv2 test (transfer from KITTI) | RMSE: 0.310 -> 0.240 | Improved |
| use quasi-random instead of random sampling on original sparse maps (R2R) (ablation) | Ours GNN · Matterport3D (transfer from NYUDv2, R2R) | RMSE: 1.831 -> 1.913 | Hurt |
| use Golden instead of random sampling on resampled maps (Q2Q) (ablation) | Ours GNN · Matterport3D (transfer from NYUDv2, Q2Q) | RMSE: 0.501 -> 0.489 | Improved |
| add GNN to baseline on KITTI (ablation) | Ours baseline · KITTI validation | RMSE (mm): 883 -> 831 | Improved |
| add GNN to baseline Rel on NYUDv2 with Golden (ablation) | Ours Baseline · NYUDv2 test | Rel: 0.016 -> 0.013 | Improved |
Closest papers by meaning
- Deep Depth Completion from Extremely Sparse Data: A Survey (TPAMI 2022 · similarity 0.90)
- GraphCSPN: Geometry-Aware Depth Completion via Dynamic GCNs (ECCV 2022 · similarity 0.90)
- Non-Local Spatial Propagation Network for Depth Completion (ECCV 2020 · similarity 0.89)
- Estimating Depth from RGB and Sparse Sensing (ECCV 2018 · similarity 0.89)
- Deep Cost Ray Fusion for Sparse Depth Video Completion (ECCV 2024 · similarity 0.89)
- Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior (AAAI 2025 · 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.
