Research map / Detection and segmentation / CNN features
Maximizing Asynchronicity in Event-based Neural Networks
Haiqing Hao, Nikola Zubic, Weihua He, Zhipeng Sui, Davide Scaramuzza, Wenhui Wang
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 temporal embedding (ablation) | EVA encoder · DVS128-Gesture (val) | SA: 81.1% -> 94.7% | Improved |
| Add temporal embedding (ablation) | EVA encoder · DVS128-Gesture (val) | FVA: 87.8% -> 98.1% | Improved |
| Add multi-vector hidden state (MVHS) (ablation) | EVA encoder · DVS128-Gesture (val) | SA: 94.1% -> 94.7% | Improved |
| Add MVHS (ablation) | EVA encoder · DVS128-Gesture (val) | MRP EC loss: 0.826 -> 0.366 | Improved |
| Add NRP objective to MRP (ablation) | EVA SSL · DVS128-Gesture (val) | SA: 94.4% -> 94.7% | Improved |
| Add TS target to MRP-EC only (ablation) | EVA SSL · DVS128-Gesture (val) | SA: 91.7% -> 94.4% | Improved |
| Learn EC, TS and NRP jointly rather than a single representation (ablation) | EVA SSL · DVS128-Gesture (val) | MRP EC pretraining loss: 0.701 -> 0.366 | Improved |
| Reduce patch size from 128 (no patching) to 16 (ablation) | EVA encoder · DVS128-Gesture (val) | SA: 89.3% -> 94.7% | Improved |
| Reduce patch size from 32 to 16 (ablation) | EVA encoder · DVS128-Gesture (val) | SA: 91.0% -> 94.7% | Improved |
| Reduce patch size from 128 to 16 (ablation) | EVA encoder · DVS128-Gesture (val) | MRP EC pretraining loss: 0.108 -> 0.366 | Hurt |
| Use EVA instead of ALERT-Tr. (+RM) | A2S encoder · DVS128-Gesture | SA: 84.6% -> 92.9% | Improved |
| Halve downsampled sequence length from 8192 to 512 | EVA · DVS128-Gesture | SA: 92.9 -> 88.4 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FARSE-CNN — "FARSE-CNN: Fully Asynchronous, Recurrent and Sparse Event-Based CNN" (ECCV 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Event-based Asynchronous Sparse Convolutional Networks (ECCV 2020 · similarity 0.91)
- Leveraging Asynchronous Spiking Neural Networks for Ultra Efficient Event-Based Visual Processing (AAAI 2025 · similarity 0.87)
- EventGPT: Event Stream Understanding with Multimodal Large Language Models (CVPR 2025 · similarity 0.86)
- PASS: Path-selective State Space Model for Event-based Recognition (NeurIPS 2025 · similarity 0.86)
- Graph Neural Network Combining Event Stream and Periodic Aggregation for Low-Latency Event-based Vision (CVPR 2025 · similarity 0.86)
- Talk2Event: Grounded Understanding of Dynamic Scenes from Event Cameras (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.
