This project demonstrates a lightweight edge ML inference pipeline focused on measuring latency and memory usage, simulating DSP-style constraints commonly found in automotive and embedded systems.
- YOLOv5 Nano (Ultralytics)
- FP16 precision
- End-to-end inference execution
- Latency measurement
- Memory footprint analysis
- Edge-oriented optimization mindset
- Latency: ~2000 ms (CPU + framework overhead)
- Memory: ~14 MB
Note: Absolute latency on CPU is not the focus. The goal is understanding bottlenecks and how DSP/accelerator-based execution would improve performance and power efficiency.
python scripts/run_inference.py