Research map / 3D vision / Depth and stereo
Multispectral Transfer Network: Unsupervised Depth Estimation for All-Day Vision
Namil Kim, Yukyung Choi, Soonmin Hwang, In So Kweon
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 chromaticity prediction as an auxiliary task (multi-task) (ablation) | STN depth network · day 1-80m | RMS: 10.418 -> 9.3808 | Improved |
| replace feature-sharing multi-task (DsMTN) with Interleaver (MTN) (ablation) | multi-task depth network · day 1-80m | RMS: 9.3808 -> 8.7387 | Improved |
| increase feature-sharing depth from low (LsMTN) to deep (DsMTN) (ablation) | multi-task depth network · day 1-80m | RMS: 9.5155 -> 9.3808 | Improved |
| add photometric correction (ablation) | MTN · day 1-80m | RMS: 10.5755 -> 8.7387 | Improved |
| add photometric correction (ablation) | MTN · night, xi=10 | ordinal accuracy: 0.3096 -> 0.4666 | Improved |
| add chromaticity multi-task and Interleaver (ablation) | STN · night, xi=10 | ordinal accuracy: 0.3233 -> 0.4666 | Improved |
| use thermal input instead of RGB input | STN · night, xi=10 | ordinal accuracy: 0.2508 -> 0.3233 | Improved |
| replace Interleaver MTN with DIW baseline at looser thresholds | MTN · night, xi=20 | ordinal accuracy: 0.7026 vs 0.7459 | Hurt |
| use MTN vs DsMTN at xi=30 (ablation) | MTN · night, xi=30 | ordinal accuracy: 0.7797 vs 0.7757 | No clear effect |
Closest papers by meaning
- Deep Depth Estimation From Thermal Image (CVPR 2023 · similarity 0.87)
- Progressive Fusion for Unsupervised Binocular Depth Estimation Using Cycled Networks (TPAMI 2019 · similarity 0.86)
- Deep Material-Aware Cross-Spectral Stereo Matching (CVPR 2018 · similarity 0.85)
- Deep 3D Capture: Geometry and Reflectance From Sparse Multi-View Images (CVPR 2020 · similarity 0.85)
- HyperDepth: Learning Depth From Structured Light Without Matching (CVPR 2016 · similarity 0.85)
- PX-NET: Simple and Efficient Pixel-Wise Training of Photometric Stereo Networks (ICCV 2021 · 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.
