Research map / Video / Video segmentation and tracking
Unsupervised Event-Based Learning of Optical Flow, Depth, and Egomotion
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 |
|---|---|---|---|
| Use event-based flow learning with discretized event volumes and motion-compensation loss instead of EV-FlowNet | Optical flow network · MVSEC outdoor_day1 and indoor_flying, dt=1 and dt=4 frames | AEE: 0.49 -> 0.32; 1.03 -> 0.58; 1.72 -> 1.02; 1.53 -> 0.87; 1.23 -> 1.30; 2.25 -> 2.18; 4.05 -> 3.85; 3.45 -> 3.18 | Mixed |
| Use event-based flow learning instead of UnFlow | Optical flow network · MVSEC, dt=1 and dt=4 frames | AEE: 0.97 -> 0.32; 0.50 -> 0.58; 0.70 -> 1.02; 0.55 -> 0.87; 2.95 -> 1.30; 3.81 -> 2.18; 6.22 -> 3.85; 1.96 -> 3.18 | Mixed |
| Use event-based flow learning instead of ECN masked | Optical flow network · MVSEC dt=1 frame | AEE: 0.36 -> 0.32; 0.20$^*$ -> 0.58; 0.24$^*$ -> 1.02; 0.21$^*$ -> 0.87 | Mixed |
| Use event-based depth learning instead of Monodepth | Depth network · MVSEC outdoor_day1, outdoor_night1, outdoor_night2, outdoor_night3 | average depth error (m): 2.72 vs 3.44; 3.84 vs 7.02; 4.40 vs 10.03; 3.13 vs 3.49; 4.02 vs 6.33; 4.89 vs 9.31; 2.19 vs 5.15; 3.15 vs 7.8; 3.92 vs | Helped |
| Use event-based egomotion learning with stereo training instead of monocular SFM Learner | Egomotion network · MVSEC outdoor_day1 | ARPE (deg): 16.27 -> 7.74 | Helped |
| Use event-based egomotion learning with stereo training instead of monocular SFM Learner | Egomotion network · MVSEC outdoor_day1 | ARRE (rad): 0.00939 -> 0.00867 | Helped |
| Use 2D convolutions over temporal channels rather than full 3D convolutions (ablation) | Event-volume network · Event motion prediction | performance | No clear effect |
| Use the bidirectional temporal loss at both event-window endpoints rather than a single target time (ablation) | Motion-compensation loss · Event motion prediction | gradient contribution across events | Helped |
Closest papers by meaning
- Unsupervised 3d Motion Estimation Using Event Camera (CVPR 2026 · similarity 0.90)
- Stereo Depth From Events Cameras: Concentrate and Focus on the Future (CVPR 2022 · similarity 0.88)
- Learning Event-Based Motion Deblurring (CVPR 2020 · similarity 0.88)
- Event-Driven Dynamic Scene Depth Completion (NeurIPS 2025 · similarity 0.87)
- E-MoFlow: Learning Egomotion and Optical Flow from Event Data via Implicit Regularization (NeurIPS 2025 · similarity 0.87)
- DERD-Net: Learning Depth from Event-based Ray Densities (NeurIPS 2025 · 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.