Research map / 3D vision / Depth and stereo
Self-Supervised Monocular Scene Flow Estimation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the 3D point reconstruction loss to the basic 2D photometric and smoothness losses (ablation) | Self-Mono-SF · KITTI split | SF1-all: 64.05 -> 49.27 | Helped |
| Apply the photometric and point reconstruction losses only to visible pixels (ablation) | Self-Mono-SF with 3D point loss · KITTI split | SF1-all: 49.27 -> 47.05 | Helped |
| Use geometric augmentation without CAM-Convs (ablation) | Self-Mono-SF · KITTI split | SF1-all: 49.89 -> 47.05 | Helped |
| Add CAM-Convs without geometric augmentation (ablation) | Self-Mono-SF · KITTI split | SF1-all: 49.89 -> 54.83 | Hurt |
| Add CAM-Convs alongside geometric augmentation (ablation) | Self-Mono-SF · KITTI split | SF1-all: 47.05 -> 49.12 | Hurt |
| Replace the single joint decoder with separate task decoders (ablation) | Self-Mono-SF · KITTI Scene Flow Training | SF1-all: 47.05 -> 100 | Hurt |
| Split the shared context network (ablation) | Self-Mono-SF · KITTI Scene Flow Training | SF1-all: 47.05 -> 62.45 | Hurt |
| Semi-supervised fine-tuning with labeled KITTI Scene Flow Training data (ablation) | Self-Mono-SF · KITTI Scene Flow Test | SF1-all: 49.54 -> 33.88 | Helped |
| Increase the 3D point reconstruction loss weight (ablation) | Self-Mono-SF · KITTI Scene Flow Training | SF1-all: The table reports 47.46, 46.08, and 46.95 for different settings. | Mixed |
| Vary the 3D smoothness loss weight (ablation) | Self-Mono-SF · KITTI Scene Flow Training | SF1-all: 62.97; 63.47; 62.55 | No clear effect |
Closest papers by meaning
- Self-Supervised Multi-Frame Monocular Scene Flow (CVPR 2021 · similarity 0.94)
- Zero-Shot Monocular Scene Flow Estimation in the Wild (CVPR 2025 · similarity 0.90)
- SelFlow: Self-Supervised Learning of Optical Flow (CVPR 2019 · similarity 0.90)
- DeltaFlow: An Efficient Multi-frame Scene Flow Estimation Method (NeurIPS 2025 · similarity 0.88)
- Flow-NeRF: Joint Learning of Geometry, Poses, and Dense Flow within Unified Neural Representations (CVPR 2025 · similarity 0.88)
- Semantic Flow: Learning Semantic Fields of Dynamic Scenes from Monocular Videos (ICLR 2024 · similarity 0.88)
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.