Research map / Architectures and efficiency / RNNs and normalization
A Chaotic Dynamics Framework Inspired by Dorsal Stream for Event Signal Processing
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Encodes event streams with a chaotic continuous-coupled neural network (CCNN) plus continuous wavelet transform into event frames, then feeds them to a ResNet-34 classifier for event-based object classification.
Problem
End-to-end deep event representations depend heavily on data structure and generalize poorly across tasks and datasets, so stable, generalizable event representations are needed for event camera object classification.
Approach
Uses a CCNN inspired by visual cortex to map constant-polarity pixel sequences to periodic outputs and varying-polarity sequences to chaotic outputs, then applies CWT and summed-real-part thresholding to produce event frames for a standard CNN classifier.
Key results
- Reports 84.4% accuracy on N-Caltech101 and 99.9% on N-CARS, outperforming the listed baselines on those two datasets.
- Achieves 97.4% on N-MNIST and 99.2% on ASL-DVS, competitive but below the best baselines (TOKEN 99.9, TORE 99.9 on ASL-DVS).
- Average inference time 2.1 ms per N-CARS sample (about 472 samples/s) on an RTX 4060 workstation.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use the chaotic dynamics event representation framework with ResNet-34 | Event-camera object classification · N-Caltech101 | accuracy: 84.4 vs. 81.6 | Helped |
| Use the chaotic dynamics event representation framework with ResNet-34 | Event-camera object classification · N-CARS | accuracy: 99.9 vs. 94.5 | Helped |
| Use the chaotic dynamics event representation framework with ResNet-34 | Event-camera object classification · ASL-DVS | accuracy: 99.2 vs. 99.9 | Hurt |
| Use the chaotic dynamics event representation framework with ResNet-34 | Event-camera object classification · N-MNIST | accuracy | Mixed |
| Use the chaotic dynamics event representation instead of Time Surface | Event representation · N-Caltech101, 30000 events | IoU: 0.4845 -> 0.4879 | Helped |
| Use the chaotic dynamics event representation instead of Time Surface | Event representation · N-Caltech101, 50000 events | IoU: 0.5976 -> 0.6087 | Helped |
| Use the chaotic dynamics event representation instead of Time Surface | Event representation · N-Caltech101, 70000 events | IoU: 0.6089 -> 0.6357 | Helped |
| Use the chaotic dynamics event representation instead of Time Surface | Event representation · N-Caltech101, 100000 events | IoU: 0.6198 -> 0.6450 | Helped |
| Use the chaotic dynamics event representation | Event-camera classification inference · N-Cars | inference time: 7.1 -> 2.1 ms | Helped |
Surprising findings
- On N-MNIST the proposed method (97.4%) is below several baselines such as TOKEN and VMV-GCN, which the paper attributes to lower dataset complexity.
Limitations
- Relies on a fixed set of hyperparameters (wavelet scale range, CCNN parameters) chosen empirically.
- Underperforms the best baselines on N-MNIST and ASL-DVS.
Experimental setup
Datasets: N-MNIST N-Caltech101 N-CARS ASL-DVS
Benchmarks: N-MNIST N-Caltech101 N-CARS ASL-DVS
Metrics: classification accuracy IoU parameters (M) MACs (G) inference time (ms)
Baselines: NDA VPT-STS GIN EventNet RG-CNNs EV-VGCNN VMV-GCN TORE HATS EST AMAE M-LSTM
Builds on: CCNN (continuous-coupled neural network) ResNet
Closest papers by meaning
- Leveraging Asynchronous Spiking Neural Networks for Ultra Efficient Event-Based Visual Processing (AAAI 2025 · similarity 0.90)
- SpikePoint: An Efficient Point-based Spiking Neural Network for Event Cameras Action Recognition (ICLR 2024 · similarity 0.86)
- Continuous Spatiotemporal Events Decoupling through Spike-based Bayesian Computation (NeurIPS 2024 · similarity 0.86)
- Efficient Event-Based Object Detection: A Hybrid Neural Network with Spatial and Temporal Attention (CVPR 2025 · similarity 0.86)
- SFOD: Spiking Fusion Object Detector (CVPR 2024 · similarity 0.85)
- PASS: Path-selective State Space Model for Event-based Recognition (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.