Research map / Video / Video segmentation and tracking
Jointly learning visual motion and confidence from local patches in event cameras
Daniel R. Kepple, Daewon Lee, Colin Prepsius, Volkan Isler, Il Memming Park, Daniel D. Lee
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 the learned confidence network (instead of heuristic event-count confidence) (ablation) | Ours · DVS-COCO test | Global MSE (deg/s)^2: 47.1 -> 4.5 | Improved |
| add the learned confidence network (instead of heuristic event-count confidence) (ablation) | Ours · DVS-COCO test | Local MSE (deg/s)^2: 48.52 -> 6.1 | Improved |
| add the local loss Lf on top of the global loss Lg (Mixture of Inputs) (ablation) | Ours · DVS-COCO test | Local MSE (deg/s)^2: 54.05 -> 6.1 | Improved |
| add the local loss Lf on top of the global loss Lg (Mixture of Inputs) (ablation) | Ours · DVS-COCO test | Global MSE (deg/s)^2: 11.6 -> 4.5 | Improved |
| add a second time constant (tau_fast and tau_slow instead of a single tau) (ablation) | Ours · DVS-COCO test | MSE (deg/s)^2: about 700 -> 4.5 | Improved |
| use local network predictions weighted by our confidence versus global contrast maximization | Ours · DVS-COCO test | Global MSE (deg/s)^2: 6.0 -> 4.5 | Improved |
| replace our local flow predictions by local contrast maximization weighted by our confidence | Contrast Maximization (local) · DVS-COCO test | Local MSE (deg/s)^2: 6.1 vs 193.7 | Hurt |
| weight local CM predictions by variance instead of mean | Contrast Maximization (local) · DVS-COCO test | Global MSE (deg/s)^2: 14.9 -> 10.4 | Improved |
| use our network confidence weights for local CM instead of variance weights | Contrast Maximization (local) · DVS-COCO test | Global MSE (deg/s)^2: 10.4 -> 28.6 | Hurt |
| our method versus Stoffregen on multiple objects sequence | Ours · EED multiple objects | success rate (%): 96.77 vs 93.3 | Hurt |
| our method versus Stoffregen on lighting variation sequence | Ours · EED lighting variation | success rate (%): 80.51 -> 97.40 | Improved |
| our network versus EV-FlowNet at 240x180 resolution | Ours · inference latency 240x180 | latency (ms): 65ms -> 12ms | Improved |
Closest papers by meaning
- Learning Visual Motion Segmentation Using Event Surfaces (CVPR 2020 · similarity 0.88)
- Event-Based Motion Segmentation by Motion Compensation (ICCV 2019 · similarity 0.88)
- The Right Spin: Learning Object Motion from Rotation-Compensated Flow Fields (IJCV 2023 · similarity 0.87)
- Unsupervised Event-Based Learning of Optical Flow, Depth, and Egomotion (CVPR 2019 · similarity 0.87)
- Learning Motion Patterns in Videos (CVPR 2017 · similarity 0.87)
- MotionMAR: Multi-scale Auto-Regressive Human Motion Reconstruction from Sparse Observations (ICML 2026 · similarity 0.87)
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.
