PolyLite RadarNet

Project Structure

PolyLite RadarNet/
├── checkpoints/           # Saved model checkpoints
├── data/                  # Data processing and loading
│   └── dataset.py         # Dataset implementation
├── log/                   # Training logs
├── models/                # Model architecture
│   ├── base_modules.py    # Basic building blocks
│   └── slowfast_base.py   # SlowFast network implementation
├── result/                # Evaluation results
├── main.py                # Training and evaluation scripts
└── readme.md              # This file

Requirements

Please see requirements.txt for a complete list of dependencies.

Installation

1. Clone this repository:

git clone https://github.com/MagicalLiHua/PolyLite-RadarNet.git
cd PolyLite-RadarNet

2. Create a virtual environment (recommended):

python -m venv venv
source venv/bin/activate  # Linux/Mac
# or
.\venv\Scripts\activate  # Windows

3. Install dependencies:

pip install -r requirements.txt

Usage

Data Preparation

Download Example Data

1. Download the example dataset from our shared drive:

   • Download Link

2. Extract the downloaded data:

unzip dataset.zip -d ./data/

Data Structure

The dataset should be organized as follows:

data/
├── datasets/
│   ├── class1/
│   │   ├── 1.npy
│   │   └── 2.npy
│   └── class2/
│       ├── 1.npy
│       └── 2.npy

Training And Evaluation

To train and evaluation the model from scratch:

python main.py

Model Architecture

The implementation is based on the SlowFast Networks architecture, which uses:

• A Slow pathway capturing spatial semantics
• A Fast pathway capturing motion dynamics
• Lateral connections between pathways

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• The implementation is based on the original SlowFast Networks paper
• Thanks to the PyTorch team for their excellent deep learning framework