Research map / 3D vision / Depth and stereo
Attentive and Contrastive Learning for Joint Depth and Motion Field Estimation
Seokju Lee, Francois Rameau, Fei Pan, 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 DAM attention module in the motion encoder (ablation) | Baseline, phase-1 · KITTI Eigen (train K) | AbsRel: 0.126 -> 0.123 | Improved |
| add DAM attention module in the motion encoder (ablation) | Baseline, phase-1 · Cityscapes (train C) | AbsRel: 0.128 -> 0.127 | Improved |
| add CSAC regularization of residual motion (ablation) | DAM, phase-3 · KITTI Eigen (train K) | AbsRel: 0.120 -> 0.114 | Improved |
| add CSAC regularization of residual motion (ablation) | DAM, phase-3 · KITTI Eigen (train C+K) | AbsRel: 0.116 -> 0.111 | Improved |
| add CSAC regularization of residual motion (ablation) | DAM, phase-3 · Cityscapes (train C) | AbsRel: 0.124 -> 0.116 | Improved |
| add CSAC regularization of residual motion (ablation) | DAM, phase-3 · Waymo Open (train W) | AbsRel: 0.155 -> 0.148 | Improved |
| add DAM after every ResBlock vs only the last ResBlock (ablation) | MotionNet encoder · KITTI phase-1 all pixels | AbsRel: 0.122 -> 0.121 | Improved |
| add DAM after every ResBlock vs only the last ResBlock (ablation) | MotionNet encoder · KITTI phase-3 all pixels | AbsRel: 0.111 -> 0.109 | No clear effect |
| use alpha 30 beta 0.2 instead of alpha 50 beta 0.1 for inlier score (ablation) | CSAC · KITTI Scene Flow 2015 foreground | D1: 34.1 -> 32.5 | Improved |
| use alpha 30 beta 0.2 instead of alpha 10 beta 0.5 for inlier score (ablation) | CSAC · KITTI Scene Flow 2015 foreground | D2: 37.2 -> 35.7 | Improved |
| reparameterize edge-aware smoothness with tau 0.1 instead of 1.0 (ablation) | motion smoothness · Cityscapes-VIS | mean IoU: 0.683 -> 0.712 | Improved |
| lower tau further from 0.1 to 0.05 (ablation) | motion smoothness · Cityscapes-VIS | mean IoU: 0.712 -> 0.704 | Hurt |
Closest papers by meaning
- Unified Depth Prediction and Intrinsic Image Decomposition from a Single Image via Joint Convolutional Neural Fields (ECCV 2016 · similarity 0.84)
- High-Rate Monocular Depth Estimation via Cross Frame-Rate Collaboration of Frames and Events (IJCV 2025 · similarity 0.84)
- Towards Robust Monocular Depth Estimation: A New Baseline and Benchmark (IJCV 2024 · similarity 0.83)
- Learning Depth from Single Monocular Images Using Deep Convolutional Neural Fields (TPAMI 2016 · similarity 0.83)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.82)
- DepthFM: Fast Generative Monocular Depth Estimation with Flow Matching (AAAI 2025 · similarity 0.82)
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.
