Research map / 3D vision / Depth and stereo
Gaussian Belief Propagation Network for Depth Completion
Jie Tang, Pingping Xie, Jian Li, Ping Tan
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 probability-based loss to the optimization baseline (ablation) | V1 baseline · NYUv2 | RMSE (mm): 342.70 -> 340.84 | Improved |
| add MRF with fixed local edges solved by serial GBP (ablation) | V2 · NYUv2 | RMSE (mm): 340.84 -> 108.29 | Improved |
| use attention-only backbone instead of conv (ablation) | V3 · NYUv2 | RMSE (mm): 108.29 -> 109.54 | Hurt |
| combine conv and attention in global-local unit (ablation) | V3 · NYUv2 | RMSE (mm): 108.29 -> 107.01 | Improved |
| add dynamic MRF parameters (ablation) | V5 · NYUv2 | RMSE (mm): 107.01 -> 103.92 | Improved |
| add dynamic non-local edges with parallel propagation (ablation) | V6 · NYUv2 | RMSE (mm): 103.92 -> 101.42 | Improved |
| increase local edges from 4 to 8 (ablation) | V7 · NYUv2 | RMSE (mm): 101.42 -> 100.95 | Improved |
| increase GBP iterations from 3 to 5 (ablation) | V8 · NYUv2 | RMSE (mm): 100.95 -> 100.69 | Improved |
| add dynamic non-local edges (delta 1.02) (ablation) | V6 · NYUv2 | delta 1.02 (%): 84.69 -> 85.19 | Improved |
| add serial GBP over MRF (delta 1.02) (ablation) | V2 · NYUv2 | delta 1.02 (%): 25.23 -> 83.71 | Improved |
| filter lowest 1% confidence pixels using predicted precision | GBPN · NYUv2 | RMSE (mm): 85.51 mm -> 26.89 mm | Improved |
| GBPN versus BP-Net in zero-shot VOID 150 points | GBPN · VOID 150 points (zero-shot from NYUv2) | RMSE (m): 0.847 -> 0.830 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OGNI-DC — OGNI-DC: Robust Depth Completion with Optimization-Guided Neural Iterations (ECCV 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CSPN++: Learning Context and Resource Aware Convolutional Spatial Propagation Networks for Depth Completion (AAAI 2020 · similarity 0.90)
- Scale Propagation Network for Generalizable Depth Completion (TPAMI 2024 · similarity 0.89)
- GraphCSPN: Geometry-Aware Depth Completion via Dynamic GCNs (ECCV 2022 · similarity 0.89)
- Uncertainty-Aware CNNs for Depth Completion: Uncertainty from Beginning to End (CVPR 2020 · similarity 0.89)
- Learning Depth with Convolutional Spatial Propagation Network (TPAMI 2019 · similarity 0.88)
- Bilateral Propagation Network for Depth Completion (CVPR 2024 · 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.
