zhengnaichuan2022

zhengnaichuan2022 / PAS-Net

Public

Official implementation of "PAS-Net: A Physics-Aware Spiking Neural Network for wearable IMU-based HAR

19
1
100% credibility
Found Apr 17, 2026 at 19 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

PAS-Net is a unified codebase for training spiking neural networks and artificial neural network baselines on multiple IMU-based human activity recognition benchmarks.

How It Works

1
🔍 Discover PAS-Net

You stumble upon this helpful project while searching for ways to recognize daily activities from wearable sensors like fitness trackers.

2
📥 Get the files

Download the ready-to-use collection of smart activity recognition tools to your computer.

3
📂 Add movement data

Place your sensor data files, like recordings from phone or watch accelerometers, into the prepared folders.

4
⚙️ Pick your style

Choose between energy-saving brain-like models or standard ones, matching your needs for quick learning or low power.

5
▶️ Start learning

Hit go, and watch the tool train itself on your data, improving at identifying walking, running, or sitting.

6
📊 Check progress

See live updates on accuracy, with logs showing how well it recognizes different movements.

✅ Perfect predictions

Enjoy spot-on activity detection from wearables, ready for health apps, fitness tracking, or research insights.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 19 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is PAS-Net?

PAS-Net delivers a Python-based framework built on PyTorch and SpikingJelly for training physics-aware spiking neural networks on wearable IMU data for human activity recognition. It tackles energy-hungry models on battery-limited devices via physics-informed tokenization from IMU signals, adaptive spiking topology, causal neuromodulation, and temporal early-exit for real-time inference. Users run one command like `python train.py --config snn-config/pamap2/pas_net.yaml` to benchmark across 8 HAR datasets with 7 ANN and 11 SNN baselines, getting accuracy-energy proxies in logs.

Why is it gaining traction?

This official GitHub repository stands out with a single unified pipeline—no separate scripts for datasets or models—plus optional compute-energy reporting (op counts times pJ/MAC assumptions) and early-exit analytics per timestep. Developers grab it for reproducible subject-independent splits on benchmarks like PAMAP2 or Opportunity, beating fragmented baselines. Official GitHub releases page ensures clean configs for quick forks.

Who should use this?

Neuromorphic engineers optimizing SNNs for smartwatches or AR glasses, wearable ML devs chasing accuracy-energy Pareto fronts on HAR tasks like locomotion or FOG detection, researchers reproducing IMU benchmarks with ANN/SNN fairness.

Verdict

Grab it as the official implementation for green HAR research—thorough README and model zoo shine despite 19 stars and 1.0% credibility score signaling early maturity. Polish for production; ideal if wearables demand spike-efficient nets over netflix pass gimmicks.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.