Research map / 3D vision / Depth and stereo
GraphCSPN: Geometry-Aware Depth Completion via Dynamic GCNs
Xin Liu, Xiaofei Shao, Bo Wang, Yali Li, Shengjin Wang
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 edge attention (ablation) | GraphCSPN · NYU-Depth-v2 (500 samples) | RMSE: 0.101 -> 0.090 | Improved |
| add geometry-aware neighborhood estimation (ablation) | GraphCSPN · NYU-Depth-v2 (500 samples) | RMSE: 0.113 -> 0.090 | Improved |
| add dynamic graph construction (ablation) | GraphCSPN · NYU-Depth-v2 (500 samples) | RMSE: 0.108 -> 0.090 | Improved |
| use L1 instead of smooth L1 loss (ablation) | GraphCSPN · NYU-Depth-v2 (500 samples) | RMSE: 0.102 -> 0.094 | Improved |
| use L1 instead of L2 loss (ablation) | GraphCSPN · NYU-Depth-v2 (500 samples) | RMSE: 0.104 -> 0.094 | Improved |
| add auxiliary loss on intermediate propagation outputs (ablation) | GraphCSPN · NYU-Depth-v2 (500 samples) | RMSE: 0.094 -> 0.090 | Improved |
| propagate from learned initial depth instead of raw input (ablation) | GraphCSPN · NYU-Depth-v2 (500 samples) | RMSE: 0.134 -> 0.090 | Improved |
| increase sparse input samples from 200 to 500 (ablation) | GraphCSPN · NYU-Depth-v2 (500 samples) | RMSE: 0.119 -> 0.090 | Improved |
| increase sparse input samples from 500 to 800 (ablation) | GraphCSPN · NYU-Depth-v2 (500 samples) | RMSE: 0.090 -> 0.082 | Improved |
| use 3 graph steps vs NLSPN 3 steps | GraphCSPN vs NLSPN · NYU-Depth-v2 (500 samples) | RMSE: 0.119 -> 0.090 | Improved |
| use 3 graph steps vs NLSPN 3 steps | GraphCSPN vs NLSPN · KITTI (Eigen split, 3200 test images) | RMSE: 2.697 -> 2.267 | Improved |
| compare GraphCSPN with NLSPN on test server | GraphCSPN · KITTI Depth Completion test | RMSE: 741.68 -> 738.41 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports GraphCSPN as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- RDFC-GAN: RGB-Depth Fusion CycleGAN for Indoor Depth Completion (TPAMI 2024 · better on 4 of 5 shared results)
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.
- DeepLiDAR — DeepLiDAR: Deep Surface Normal Guided Depth Prediction for Outdoor Scene From Sparse LiDAR Data and Single Color Image (CVPR 2019) · better on 1 of 1 shared result
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 0.93)
- Bilateral Propagation Network for Depth Completion (CVPR 2024 · similarity 0.90)
- DeCoTR: Enhancing Depth Completion with 2D and 3D Attentions (CVPR 2024 · similarity 0.88)
- Efficient Indoor Depth Completion Network Using Mask-adaptive Gated Convolution (AAAI 2025 · similarity 0.87)
- Masked Spatial Propagation Network for Sparsity-Adaptive Depth Refinement (CVPR 2024 · similarity 0.87)
- Completion as Enhancement: A Degradation-Aware Selective Image Guided Network for Depth Completion (CVPR 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.
