Research map / 3D vision / Depth and stereo
SemARFlow: Injecting Semantics into Unsupervised Optical Flow Estimation for Autonomous Driving
Shuai Yuan, Shuzhi Yu, Hannah Kim, Carlo Tomasi
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 semantic segmentation inputs, encoder and augmentation, to adapted ARFlow | ARFlow · KITTI-2015 test | Fl-all (%): 11.80 -> 8.38 | Improved |
| add semantic encoder to baseline (ablation) | SemARFlow baseline · KITTI-2015 test | Fl-all (%): 9.89 -> 8.99 | Improved |
| add semantic augmentation on top of semantic encoder (ablation) | SemARFlow · KITTI-2015 test | Fl-all (%): 8.99 -> 8.38 | Improved |
| add learned upsampler while keeping smoothness loss (ablation) | ARFlow · KITTI-2015 train | Fl-all (%): 10.36 -> 10.22 | No clear effect |
| turn off smoothness loss together with learned upsampler (ablation) | ARFlow with upsampler · KITTI-2015 train | Fl-all (%): 10.22 -> 8.87 | Improved |
| add semantic encoder at level 3 versus other levels (ablation) | ARFlow · KITTI-2015 train | Fl-all (%): 10.36 -> 9.75 | Improved |
| add semantic encoder on upsampler/no-smoothness model (ablation) | SemARFlow · KITTI-2015 train | Fl-all (%): 8.87 -> 8.26 | Improved |
| add semantic augmentation on upsampler/no-smoothness model (ablation) | SemARFlow · KITTI-2015 train | Fl-all (%): 8.87 -> 8.80 | Improved |
| start semantic augmentation at 150k rather than 100k iterations (ablation) | SemARFlow · KITTI-2015 train | Fl-all (%): 7.92 -> 7.79 | Improved |
| augment poles and sky in addition to vehicles (ablation) | SemARFlow · KITTI-2015 train | Fl-all (%): 7.94 -> 7.79 | Improved |
| apply augmentation loss on full region rather than only newly occluded region (ablation) | SemARFlow · KITTI-2015 train | Fl-all (%): 8.15 -> 7.79 | Improved |
| add semantic inputs for cross-dataset training | SemARFlow · Cityscapes to KITTI-2015 train | Fl-all (%): 12.27 -> 10.32 | Improved |
Closest papers by meaning
- Exploiting Semantic Information and Deep Matching for Optical Flow (ECCV 2016 · similarity 0.91)
- UnSAMFlow: Unsupervised Optical Flow Guided by Segment Anything Model (CVPR 2024 · similarity 0.91)
- M2Flow: A Motion Information Fusion Framework for Enhanced Unsupervised Optical Flow Estimation in Autonomous Driving (AAAI 2025 · similarity 0.90)
- UniFlow: Zero-Shot LiDAR Scene Flow for Autonomous Driving (ECCV 2026 · similarity 0.89)
- SAMFlow: Eliminating Any Fragmentation in Optical Flow with Segment Anything Model (AAAI 2024 · similarity 0.87)
- ICP-Flow: LiDAR Scene Flow Estimation with ICP (CVPR 2024 · similarity 0.86)
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.
