Research map / Detection and segmentation / CNN features
Depth Coefficients for Depth Completion
Saif Imran, Yunfei Long, Xiaoming Liu, Daniel Morris
Check 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 |
|---|---|---|---|
| Use Depth Coefficients with cross-entropy instead of sparse depth with MSE (ablation) | Depth completion network · NYU2 | MAE: 6.63 -> 3.82 | Improved |
| Use Depth Coefficients input instead of sparse-depth input with MSE loss (ablation) | Depth completion network · NYU2 | MAE: 6.63 -> 6.10 | Improved |
| Use cross-entropy instead of MSE with sparse-depth input (ablation) | Depth completion network · NYU2 | MAE: 6.63 -> 9.53 | Hurt |
| Use 3-coefficient reconstruction instead of all coefficients (ablation) | Depth completion network · KITTI, 16R Lidar | RMSE: 142.3 -> 160.6 | Hurt |
| Use 3-coefficient reconstruction instead of all coefficients (ablation) | Depth completion network · KITTI, 16R Lidar | tRMSE: 50.5 -> 47.2 | Improved |
| Use 3-coefficient reconstruction instead of all coefficients (ablation) | Depth completion network · KITTI, 16R Lidar | tMAE: 36.1 -> 33.4 | Improved |
| Use 3-coefficient reconstruction instead of all coefficients (ablation) | Depth completion network · NYU2 | RMSE: 0.118 -> 0.131 | Hurt |
| Use 3-coefficient reconstruction instead of all coefficients (ablation) | Depth completion network · NYU2 | tMAE: 0.042 -> 0.040 | Improved |
| Increase DC channels from 10 to 40 (ablation) | Depth completion network · KITTI validation, 16R Lidar | MAE: 50.2 -> 37.8 | Improved |
| Increase DC channels from 40 to 80 (ablation) | Depth completion network · KITTI validation, 16R Lidar | MAE: 37.8 -> 38.4 | Hurt |
| Upsample sparse 16-row Lidar with DC-3coeff instead of Ma | Car detection with Frustum PointNet · KITTI 3D bounding boxes, Easy | average precision: 36.7 -> 64.9 | Improved |
| Upsample sparse 16-row Lidar with Ma instead of using raw sparse data | Car detection with Frustum PointNet · KITTI 3D bounding boxes, Easy | average precision: 54.4 -> 36.7 | Hurt |
Closest papers by meaning
- OMNI-DC: Highly Robust Depth Completion with Multiresolution Depth Integration (ICCV 2025 · similarity 0.88)
- Deep Depth Completion of a Single RGB-D Image (CVPR 2018 · similarity 0.88)
- Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior (AAAI 2025 · similarity 0.87)
- Improving Depth Completion via Depth Feature Upsampling (CVPR 2024 · similarity 0.86)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.86)
- DeCoTR: Enhancing Depth Completion with 2D and 3D Attentions (CVPR 2024 · similarity 0.85)
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.
