Research map / 3D vision / Depth and stereo
Estimating Depth from RGB and Sparse Sensing
Zhao Chen, Vijay Badrinarayanan, Gilad Drozdov, Andrew Rabinovich
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 sparse depth input (0.18% points) to the depth network (ablation) | D3 · NYUv2 val | RMSE (m): 0.711 -> 0.118 | Improved |
| add distance map S2 to the sparse input parametrization (ablation) | D3 · validation L2 loss, random sparsity schedule | relative L2 validation loss: 34.4% improvement | Improved |
| add distance map S2 under slow-decay schedule (ablation) | D3 · validation L2 loss | relative L2 validation loss: 8.8% improvement | Improved |
| add distance map S2 under static schedule (ablation) | D3 · validation L2 loss | relative L2 validation loss: 2.8% improvement | Improved |
| use nearest-neighbor fill of sparse depth instead of raw sparse map (ablation) | D3 · NYUv2, 0.18% sparsity, standard schedule | RMSE (m): 0.287 -> 0.118 | Improved |
| use nearest-neighbor fill of sparse depth instead of raw sparse map (ablation) | D3 · NYUv2, 0.18% sparsity, random schedule | RMSE (m): 0.344 -> 0.125 | Improved |
| feed S1 into every DenseNet module (ablation) | D3 · NYUv2, standard schedule | RMSE (m): 0.120 -> 0.118 | No clear effect |
| feed S1 into every DenseNet module (ablation) | D3 · NYUv2, random schedule | RMSE (m): 0.135 -> 0.125 | Improved |
| use slow-decay sparsity schedule instead of static pattern (ablation) | D3 · training L2 loss and MRE | relative decrease in loss and MRE: ~3% | Improved |
| train jointly on NYUv2 and KITTI (mixed) | D3 · NYUv2 0.18% sparsity | RMSE (m): 0.118 -> 0.131 | Hurt |
| train jointly on NYUv2 and KITTI (mixed) | D3 · KITTI 0.174% | RMSE (m): 1.387 -> 1.472 | Hurt |
| average depth over a larger patch (11px vs 1px) for sparse samples | D3 · NYUv2 0.18% sparsity | RMSE (m): 0.118 -> 0.102 | Improved |
Closest papers by meaning
- Sparse Auxiliary Networks for Unified Monocular Depth Prediction and Completion (CVPR 2021 · similarity 0.91)
- Neural RGB(r)D Sensing: Depth and Uncertainty From a Video Camera (CVPR 2019 · similarity 0.90)
- RGB-Depth Fusion GAN for Indoor Depth Completion (CVPR 2022 · similarity 0.90)
- Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (ICLR 2025 · similarity 0.88)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.88)
- Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior (AAAI 2025 · similarity 0.87)
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.
