Research map / 3D vision / Depth and stereo
T2Net: Synthetic-to-Realistic Translation for Solving Single-Image Depth Estimation Tasks
Chuanxia Zheng, Tat-Jen Cham, Jianfei Cai
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 unpaired real images with translation and GAN losses (full T2Net vs synthetic-only) | depth network f_T · KITTI Eigen split | Abs Rel: 0.278 -> 0.169 | Improved |
| add unpaired real images with translation and GAN losses (full T2Net vs synthetic-only) | depth network f_T · NYUDv2 | Abs Rel: 0.304 -> 0.257 | Improved |
| train translation and task networks end-to-end rather than separately (ablation) | T2Net · KITTI Eigen split | Abs Rel: 0.234 -> 0.169 | Improved |
| add reconstruction loss on real images (ablation) | T2Net · KITTI Eigen split | Abs Rel: 0.201 -> 0.169 | Improved |
| add the feature-level GAN loss D_feat (ablation) | T2Net · KITTI Eigen split | Abs Rel: 0.168 -> 0.169 | No clear effect |
| add the feature-level GAN loss D_feat to the image GAN (ablation) | T2Net · NYUDv2 | Abs Rel: 0.274 -> 0.257 | Improved |
| add the image GAN loss (ablation) | T2Net · KITTI Eigen split | Abs Rel: 0.233 -> 0.169 | Improved |
| use image GAN alone instead of feature GAN alone (ablation) | T2Net · NYUDv2 | Abs Rel: 0.320 -> 0.274 | Improved |
| use target-domain reconstruction with unidirectional translation instead of separately trained CycleGAN (ablation) | T2Net · KITTI Eigen split | Abs Rel: 0.212 -> 0.169 | Improved |
| use target-domain reconstruction on real images instead of self-domain reconstruction on synthetic images (ablation) | T2Net · KITTI Eigen split | Abs Rel: 0.199 -> 0.169 | Improved |
| use the proposed task network instead of a vanilla task network (synthetic only) (ablation) | f_T · KITTI Eigen split | Abs Rel: 0.295 -> 0.278 | Improved |
| use the proposed task network instead of a vanilla task network (full approach) (ablation) | T2Net · KITTI Eigen split | Abs Rel: 0.259 -> 0.169 | Improved |
Closest papers by meaning
- Deep CG2Real: Synthetic-to-Real Translation via Image Disentanglement (ICCV 2019 · similarity 0.88)
- Domain Decluttering: Simplifying Images to Mitigate Synthetic-Real Domain Shift and Improve Depth Estimation (CVPR 2020 · similarity 0.87)
- StereoGAN: Bridging Synthetic-to-Real Domain Gap by Joint Optimization of Domain Translation and Stereo Matching (CVPR 2020 · similarity 0.87)
- Unsupervised CNN for Single View Depth Estimation: Geometry to the Rescue (ECCV 2016 · similarity 0.87)
- DepthFM: Fast Generative Monocular Depth Estimation with Flow Matching (AAAI 2025 · similarity 0.86)
- Compressing Image-to-Image Translation GANs Using Local Density Structures on Their Learned Manifold (AAAI 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.
