Research map / 3D vision / Depth and stereo
Deep Depth Completion of a Single RGB-D Image
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Predict normals and occlusion boundaries from color, then solve depth globally | RGB-D depth completion · Matterport3D test split, unobserved pixels | Rel: 0.089 vs. 0.103 | Helped |
| Use color alone instead of color and depth as network input (ablation) | Normal prediction network · Matterport3D test split, unobserved pixels | Rel: 0.090 -> 0.089 | Helped |
| Use color instead of depth as network input (ablation) | Normal prediction network · Matterport3D test split, unobserved pixels | Rel: 0.107 -> 0.089 | Helped |
| Predict surface normals and solve for depth rather than predict depth directly (ablation) | RGB-D depth completion · Matterport3D test split, unobserved pixels | Rel: 0.408 -> 0.089 | Helped |
| Predict normals rather than depth derivatives (ablation) | RGB-D depth completion · Matterport3D test split, unobserved pixels | Rel: 0.100 -> 0.089 | Helped |
| Add occlusion-boundary weighting to down-weight normal constraints near boundaries (ablation) | Depth optimization from predicted normals · Matterport3D test split, unobserved pixels | Rel: 0.110 -> 0.089 | Helped |
| Use rendered-depth targets instead of raw-depth targets (ablation) | Normal prediction network · Matterport3D test split, unobserved pixels | Rel: 0.094 -> 0.089 | Helped |
| Train with loss on both observed and unobserved pixels rather than observed pixels only (ablation) | Normal prediction network · Matterport3D test split, unobserved pixels | Rel: 0.091 -> 0.089 | Helped |
| Train with loss on both observed and unobserved pixels rather than unobserved pixels only (ablation) | Normal prediction network · Matterport3D test split, unobserved pixels | Rel: 0.090 -> 0.089 | Helped |
| Add predicted depth derivatives alongside normals (ablation) | Depth optimization from predicted geometry · Matterport3D test split, unobserved pixels | Rel: 0.089 -> 0.092 | Hurt |
Closest papers by meaning
- Neural RGB(r)D Sensing: Depth and Uncertainty From a Video Camera (CVPR 2019 · similarity 0.90)
- RGB-D Local Implicit Function for Depth Completion of Transparent Objects (CVPR 2021 · similarity 0.90)
- Completion as Enhancement: A Degradation-Aware Selective Image Guided Network for Depth Completion (CVPR 2025 · similarity 0.88)
- Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior (AAAI 2025 · similarity 0.88)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.87)
- Efficient Indoor Depth Completion Network Using Mask-adaptive Gated Convolution (AAAI 2025 · similarity 0.86)
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.