Research map / Video / Video segmentation and tracking
SuperEvent: Cross-Modal Learning of Event-based Keypoint Detection for SLAM
Yannick Burkhardt, Simon Schaefer, Stefan Leutenegger
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 |
|---|---|---|---|
| train with temporal matching pseudo-labels from real event pairs instead of homographic adaptation (ablation) | SuperEvent · ECD pose estimation | AUC@20deg: 31.0 -> 46.7 | Improved |
| mix homographic adaptation data into temporal matching training data (ablation) | SuperEvent · ECD pose estimation | AUC@20deg: 46.7 -> 37.1 | Hurt |
| use two-polarity multi-channel time surface with single window instead of Tencode (ablation) | SuperEvent · ECD pose estimation | AUC@20deg: 36.9 -> 38.7 | Improved |
| use five time windows (10-channel MCTS_5) instead of one window (ablation) | SuperEvent · ECD pose estimation | AUC@20deg: 38.7 -> 46.7 | Improved |
| replace VGG backbone by MaxViT transformer backbone (3 layers) (ablation) | SuperEvent · ECD pose estimation | AUC@20deg: 42.2 -> 46.7 | Improved |
| use pixelwise descriptor head instead of 8x8 grid descriptor (ablation) | SuperEvent (MaxViT) · ECD pose estimation | AUC@20deg: 46.7 -> 38.0 | Hurt |
| add a fourth backbone layer (ablation) | SuperEvent · ECD pose estimation | AUC@20deg: 46.7 -> 43.8 | Hurt |
| integrate SuperEvent keypoints into OKVIS2 stereo VI-SLAM | OKVIS2 · TUM-VIE mocap average | ATE (cm): 4.78 (ESVO2) -> 0.55 | Improved |
| enable OKVIS2 loop closure with SuperEvent descriptors (ablation) | OKVIS2 + SuperEvent · TUM-VIE loop-floor0 | ATE (cm): 132.11 -> 4.96 | Improved |
| enable loop closure on small mocap sequences (ablation) | OKVIS2 + SuperEvent · TUM-VIE mocap average | ATE (cm): 0.57 -> 0.55 | No clear effect |
| raise SuperEvent processing rate from 20 Hz to 40 Hz (ablation) | OKVIS2 + SuperEvent · TUM-VIE mocap-shake | ATE (cm): 43.71 -> 29.14 | Improved |
Closest papers by meaning
- SD2Event:Self-supervised Learning of Dynamic Detectors and Contextual Descriptors for Event Cameras (CVPR 2024 · similarity 0.88)
- PASS: Path-selective State Space Model for Event-based Recognition (NeurIPS 2025 · similarity 0.87)
- Learning Graph-embedded Key-event Back-tracing for Object Tracking in Event Clouds (NeurIPS 2022 · similarity 0.87)
- ETAP: Event-based Tracking of Any Point (CVPR 2025 · similarity 0.86)
- IncEventGS: Pose-Free Gaussian Splatting from a Single Event Camera (CVPR 2025 · similarity 0.85)
- FlexEvent: Towards Flexible Event-Frame Object Detection at Varying Operational Frequencies (NeurIPS 2025 · similarity 0.85)
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.
