Research map / 3D vision / Depth and stereo
LocalBins: Improving Depth Estimation by Learning Local Distributions
Shariq Farooq Bhat, Ibraheem Alhashim, Peter Wonka
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 the LocalBins module to the encoder-decoder (ablation) | base encoder-decoder · NYU-Depth V2 test (ablation) | REL: 0.111 -> 0.106 | Improved |
| add Query-Response training with Chamfer loss (ablation) | LocalBins module · NYU-Depth V2 test (ablation) | REL: 0.106 -> 0.099 | Improved |
| add Query-Response training instead of naive 5x5 training (ablation) | LocalBins · NYU-Depth V2 test (ablation) | REL: 0.107 -> 0.099 | Improved |
| add foveated weighting in Chamfer loss (ablation) | LocalBins · NYU-Depth V2 test (ablation) | RMS: 0.364 -> 0.357 | Improved |
| add foveated weighting in Chamfer loss (ablation) | LocalBins · NYU-Depth V2 test (ablation) | REL: 0.099 -> 0.099 | No clear effect |
| use naive training with larger box 3x3 to 15x15 (ablation) | LocalBins · NYU-Depth V2 test (ablation) | REL: 0.108, 0.107, 0.108 | No clear effect |
| use Query-Response with 1000 queries instead of 8192 naive points (ablation) | LocalBins · NYU-Depth V2 test (ablation) | REL: 0.1071 -> 0.0992 | Improved |
| use sigmoid splitter instead of constant splitter (ablation) | LocalBins · NYU-Depth V2 test (ablation) | REL: 0.117 -> 0.100 | Improved |
| use linear norm splitter instead of sigmoid splitter (ablation) | LocalBins · NYU-Depth V2 test (ablation) | REL: 0.100 -> 0.099 | No clear effect |
| replace AdaBins with LocalBins (same backbone) | EfficientNet-B5 · NYU-Depth V2 test | REL: 0.103 -> 0.099 | Improved |
| replace AdaBins with LocalBins zero-shot | EfficientNet-B5 · iBims-1 (zero-shot from NYU) | REL: 0.159 -> 0.156 | Improved |
| replace AdaBins with LocalBins zero-shot | EfficientNet-B5 · SUN-RGBD (zero-shot from NYU) | RMS: 0.90 -> 0.88 | Improved |
Closest papers by meaning
- AdaBins: Depth Estimation Using Adaptive Bins (CVPR 2021 · similarity 0.92)
- DME: Unveiling the Bias for Better Generalized Monocular Depth Estimation (AAAI 2024 · similarity 0.87)
- PatchFusion: An End-to-End Tile-Based Framework for High-Resolution Monocular Metric Depth Estimation (CVPR 2024 · similarity 0.87)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.86)
- Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data (CVPR 2024 · similarity 0.86)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · 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.
