How venu277/neuromorphic-event-camera-perception-for-automotive-applications works
This repository is a research pipeline for event-camera car-versus-background perception and SNN/CNN tradeoff evaluation. The documented workflow starts with N-MNIST event framing and SNN training, then parses and bins raw N-Cars events, trains spiking and conventional classifiers, and measures accuracy, spike/FLOP proxies, and CPU latency. The N-Cars parser and dataset handling are embedded in several scripts rather than factored into a shared module. The README documents the head-to-head findings; the sampled benchmark scripts measure their respective models separately.
Main entry points: step8_train_cnn_baseline.py Train car CNN · step9_benchmark_snn.py Measure SNN · step4_train_snn.py Train digit SNN · step7_train_ncars_snn.py SNN event loader
Components 15
N-MNIST Baseline
- N-MNIST dataset
- Inspect event frames
step3_load_nmnist.py - Train digit SNN
step4_train_snn.py - Digit SNN weights
snn_nmnist_model.pth
N-Cars Data
- N-Cars event files
- Inspect parsed sample
step6_load_ncars.py - SNN event loader
step7_train_ncars_snn.py - CNN event loader
step8_train_cnn_baseline.py - Benchmark event loader
step9_benchmark_snn.py
Classifier Training
- Train car SNN
step7_train_ncars_snn.py - Train car CNN
step8_train_cnn_baseline.py - Car SNN weights
snn_ncars_model_v2.pth - Car CNN weights
cnn_ncars_baseline.pth
Model Evaluation
- Measure SNN
step9_benchmark_snn.py - Tradeoff findings
README.md
Connections 15
- N-MNIST dataset → Inspect event frames: loads samples
- N-MNIST dataset → Train digit SNN: loads and frames
- Train digit SNN → Digit SNN weights: saves weights
- N-Cars event files → Inspect parsed sample: reads events
- N-Cars event files → SNN event loader: parses and bins
- N-Cars event files → CNN event loader: parses and bins
- N-Cars event files → Benchmark event loader: parses and bins
- SNN event loader → Train car SNN: supplies batches
- CNN event loader → Train car CNN: supplies batches
- Train car SNN → Car SNN weights: saves weights
- Train car CNN → Car CNN weights: saves weights
- Car SNN weights → Measure SNN: loads weights
- Benchmark event loader → Measure SNN: supplies test samples
- Train car CNN → Tradeoff findings: documents CNN results
- Measure SNN → Tradeoff findings: documents SNN results