Research map / 3D vision / Depth and stereo
CostDCNet: Cost Volume Based Depth Completion for a Single RGB-D Image
Jaewon Kam, Jungeon Kim, Soongjin Kim, Jaesik Park, Seungyong Lee
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 |
|---|---|---|---|
| replace 2D UNet with tiny 3D UNet on RGB-D feature volume (ablation) | CostDCNet · NYUv2 ablation (1/4 resolution, 20 epochs) | RMSE (m): 0.195 -> 0.121 | Improved |
| use pseudo-3D convolution instead of full 3D convolution (ablation) | 3D UNet · NYUv2 ablation (1/4 resolution, 20 epochs) | RMSE (m): 0.121 -> 0.123 | No clear effect |
| add geometric feature volume to image feature volume (ablation) | RGB-D volume Type A · NYUv2 ablation (1/4 resolution, 20 epochs) | RMSE (m): 0.186 -> 0.123 | Improved |
| assign unknown-depth features to all depth planes (Type A vs Type B) (ablation) | RGB-D volume · NYUv2 ablation (1/4 resolution, 20 epochs) | RMSE (m): 0.131 -> 0.123 | Improved |
| assign valid-depth features to correct depth planes (Type C vs Type A) (ablation) | RGB-D volume · NYUv2 ablation (1/4 resolution, 20 epochs) | RMSE (m): 0.123 -> 0.122 | Improved |
| add 2D and 3D encoders (ablation) | CostDCNet Type C · NYUv2 ablation (1/4 resolution, 20 epochs) | RMSE (m): 0.122 -> 0.108 | Improved |
| use bilateral grid learning instead of bilinear upsampling (ablation) | CostDCNet · NYUv2 ablation (1/4 resolution, 20 epochs) | RMSE (m): 0.106 -> 0.102 | Improved |
| use per-plane pixel shuffle instead of bilinear upsampling (ablation) | CostDCNet · NYUv2 ablation (1/4 resolution, 20 epochs) | RMSE (m): 0.106 -> 0.099 | Improved |
| use per-plane pixel shuffle instead of bilateral grid (ablation) | CostDCNet · NYUv2 ablation (1/4 resolution, 20 epochs) | parameters: 3.1M -> 1.8M | Improved |
| use CostDCNet instead of Huang et al. | depth completion · Matterport3D semi-dense | RMSE (m): 1.092 -> 1.019 | Improved |
| use CostDCNet instead of NLSPN with 500 points | depth completion · NYUv2 (500 points) | RMSE (m): 0.092 -> 0.096 | Hurt |
| use CostDCNet instead of Point-Fusion with 32 points | depth completion · NYUv2 (32 points) | RMSE (m): 0.319 -> 0.258 | 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.
- 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
- Cost Volume Pyramid Based Depth Inference for Multi-View Stereo (CVPR 2020 · similarity 0.92)
- Deep Depth Completion of a Single RGB-D Image (CVPR 2018 · similarity 0.90)
- Learning Joint 2D-3D Representations for Depth Completion (ICCV 2019 · similarity 0.89)
- Non-Parametric Depth Distribution Modelling Based Depth Inference for Multi-View Stereo (CVPR 2022 · similarity 0.89)
- AGG-Net: Attention Guided Gated-Convolutional Network for Depth Image Completion (ICCV 2023 · similarity 0.89)
- A Survey on Deep Learning Techniques for Stereo-Based Depth Estimation (TPAMI 2020 · similarity 0.89)
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.
