Research map / 3D vision / Depth and stereo
Learning Normal Flow Directly From Events
Dehao Yuan, Levi Burner, Jiayi Wu, Minghui Liu, Jingxi Chen, Yiannis Aloimonos, Cornelia Fermüller
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 |
|---|---|---|---|
| replace conventional optical flow loss with the two-term motion field loss (radial plus angular) (ablation) | point-based normal flow estimator · ablation (Appendix 11) | error | Improved |
| add uncertainty-score filtering from rotation ensemble (ablation) | egomotion estimation · EVIMO2-sfm | egomotion error | Improved |
| add rotation-ensemble uncertainty score (ablation) | normal flow estimator · ablation | correlation of error with uncertainty | Improved |
| use SVM maximum-margin egomotion solver instead of depth-positivity maximization | egomotion from normal flow · EVIMO2-sfm 03 02 0 | RMS V (m/s): 0.500 -> 0.227 | Improved |
| use SVM solver instead of depth-positivity method | egomotion from normal flow · EVIMO2-sfm 03 04 0 | RMS V (m/s): 0.217 -> 0.136 | Improved |
| use SVM solver instead of depth-positivity method | egomotion from normal flow · EVIMO2-sfm 03 02 3 | RMS V (m/s): 0.213 -> 0.279 | Hurt |
| use point-based VecKM estimator instead of PointNet (both supervised) | normal flow estimation · EVIMO2-imo average | PEE: 0.933 -> 0.423 | Improved |
| use point-based supervised estimator instead of frame-based self-supervised TCM | normal flow estimation · MVSEC average | PEE: 0.516 -> 0.992 | Hurt |
| use point-based estimator instead of frame-based TCM | normal flow estimation · EVIMO2-imo average | PEE: 1.105 -> 0.396 | Improved |
| train on union of MVSEC, DSEC and EVIMO instead of single dataset | normal flow estimator · MVSEC average | PEE: 1.419 -> 0.992 | Improved |
| train on DSEC single scene instead of MVSEC | normal flow estimator · DSEC 8 day scenes | PEE: 1.631 -> 1.043 | Improved |
| evaluate at night instead of day | normal flow estimator trained on DSEC · DSEC | PEE: 1.043 -> 1.357 | Hurt |
Closest papers by meaning
- Learning Efficient Meshflow and Optical Flow From Event Cameras (TPAMI 2025 · similarity 0.90)
- Secrets of Event-Based Optical Flow (ECCV 2022 · similarity 0.90)
- Event-based Optical Flow Using Spatio-temporal Registration (TPAMI 2026 · similarity 0.89)
- EDCFlow: Exploring Temporally Dense Difference Maps for Event-based Optical Flow Estimation (CVPR 2025 · similarity 0.88)
- MemFlow: Optical Flow Estimation and Prediction with Memory (CVPR 2024 · similarity 0.88)
- Efficient Meshflow and Optical Flow Estimation from Event Cameras (CVPR 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.
