Research map / 3D vision / Depth and stereo
Unsupervised CNN for Single View Depth Estimation: Geometry to the Rescue
Ravi Garg, Vijay Kumar B.G., Gustavo Carneiro, Ian Reid
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 8x augmentation fine-tuning (color, scale, flips) (ablation) | Ours L12 · KITTI (Eigen test split, 1-50m) | RMS: 5.285 -> 5.104 | Improved |
| add 8x augmentation fine-tuning (ablation) | Ours L12 · KITTI (Eigen test split, 1-50m) | absolute relative error: 0.177 -> 0.169 | Improved |
| add 8x augmentation fine-tuning (ablation) | Ours L12 · KITTI (Eigen test split, 1-50m) | accuracy delta<1.25: 0.727 -> 0.740 | Improved |
| add third skip-connection at L10 (ablation) | Ours L10 · KITTI (Eigen test split, 1-50m) | RMS: 5.434 -> 5.850 | Hurt |
| add third skip-connection at L10 (ablation) | Ours L10 · KITTI (Eigen test split, 1-50m) | absolute relative error: 0.189 -> 0.246 | Hurt |
| increase output resolution from L10 to L12 via more upsampling layers (ablation) | Ours · KITTI (Eigen test split, 1-50m) | RMS: 5.434 -> 5.285 | Improved |
| increase output resolution from L9 to L10 (ablation) | Ours · KITTI (Eigen test split, 1-50m) | RMS: 5.740 -> 5.434 | Improved |
| train with photometric reconstruction loss instead of supervising on Horn-Schunck stereo output (ablation) | CNN · KITTI (Eigen test split, 1-50m) | RMS: 6.292 -> 5.285 | Improved |
| train with photometric loss instead of supervising on SGM stereo output (ablation) | CNN · KITTI (Eigen test split, 1-50m) | RMS: 5.680 -> 5.285 | Improved |
| train with photometric loss instead of HS->CNN with gamma=.01 (ablation) | CNN · KITTI (Eigen test split, 1-50m) | accuracy delta<1.25: 0.476 -> 0.727 | Improved |
| use lower regularization gamma=0.01 for HS->CNN stereo supervision (ablation) | HS->CNN · KITTI (stereo baseline comparison) | RMS: 6.292 -> 6.691 | Hurt |
| unsupervised photometric training vs supervised Eigen et al fine | Ours L12 · KITTI (Eigen test split, 1-50m) | RMS: 7.156 -> 5.285 | Improved |
Closest papers by meaning
- Unsupervised Monocular Depth Estimation With Left-Right Consistency (CVPR 2017 · similarity 0.92)
- Single Image Depth Estimation Trained via Depth From Defocus Cues (CVPR 2019 · similarity 0.92)
- Learning Depth from Single Monocular Images Using Deep Convolutional Neural Fields (TPAMI 2016 · similarity 0.92)
- Visualization of Convolutional Neural Networks for Monocular Depth Estimation (ICCV 2019 · similarity 0.92)
- CAM-Convs: Camera-Aware Multi-Scale Convolutions for Single-View Depth (CVPR 2019 · similarity 0.90)
- Learning Depth From Monocular Videos Using Direct Methods (CVPR 2018 · similarity 0.90)
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.
