Research map / 3D vision / Depth and stereo
RADU: Ray-Aligned Depth Update Convolutions for ToF Data Denoising
Michael Schelling, Pedro Hermosilla, Timo Ropinski
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 |
|---|---|---|---|
| Add cyclic self-training U-DA | RADU · S4 | MAE [cm]: 1.83 -> 2.11 | Hurt |
| Add cyclic self-training U-DA | RADU · S5 | MAE [cm]: 2.59 -> 1.63 | Improved |
| Use RADU instead of TIR | ToF denoising · S4 | MAE [cm]: 2.60 -> 1.83 | Improved |
| Use RADU instead of CFN | ToF denoising · Cornell-Box | MAE [cm]: 3.99 -> 3.64 | Improved |
| Use RADU instead of CFN | ToF denoising · FLAT | MAE [cm]: 6.29 -> 3.31 | Improved |
| Replace standard MCConv with MCConv plus ray-aligned depth updates (ablation) | Latent 3D block · S3 validation | MAE [cm]: 2.51 -> 2.28 | Improved |
| Use 2.5D convolutions instead of 2D convolutions (ablation) | Latent block · S3 validation | MAE [cm]: 2.72 -> 2.61 | Improved |
| Remove ray-aligned point updates by setting α=0.0 (ablation) | RADU · S3 validation | MAE [cm]: 2.28 -> 2.51 | Hurt |
| Increase update scale α from 0.1 to 0.2 (ablation) | RADU · S3 validation | MAE [cm]: 2.28 -> 2.81 | Hurt |
| Increase update scale α from 0.1 to 1.0 (ablation) | RADU · S3 validation | MAE [cm]: 2.28 -> 3.42 | Hurt |
| Set update scale α to each layer's receptive-field radius (ablation) | RADU · S3 validation | MAE [cm]: 2.28 -> 2.45 | Hurt |
| Remove the intermediate coarse-depth L1 loss (ablation) | RADU · S3 | MAE [cm]: 2.28 -> 2.63; +0.35 | Hurt |
Closest papers by meaning
- Unsupervised Domain Adaptation of Deep Networks for ToF Depth Refinement (TPAMI 2021 · similarity 0.88)
- Deep End-to-End Time-of-Flight Imaging (CVPR 2018 · similarity 0.88)
- Tackling 3D ToF Artifacts Through Learning and the FLAT Dataset (ECCV 2018 · similarity 0.88)
- Unsupervised Domain Adaptation for ToF Data Denoising With Adversarial Learning (CVPR 2019 · similarity 0.87)
- Spatial Hierarchy Aware Residual Pyramid Network for Time-of-Flight Depth Denoising (ECCV 2020 · similarity 0.87)
- Deep End-to-End Alignment and Refinement for Time-of-Flight RGB-D Module (ICCV 2019 · 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.
