MAML_pth

PyTorch implementations of various classification models adapted for few-shot learning using Model-Agnostic Meta-Learning.
The model is trained on the miniImageNet dataset using class-balanced episodic tasks.

Features

• PyTorch implementations of various models, adapted for MAML
• Meta-learning framework for few-shot tasks
• Support for N-way K-shot training episodes
• Few-shot learning support
• Training on miniImageNet dataset
• Checkpoint management for model states
• Inference pipeline for predictions

Requirements

• Python 3.11+
• PyTorch 2.5+ (with CUDA 12.6)
• Kaggle API credentials

Setup

1. (Optional) Create a virtual environment:

   python -m venv venv
   source venv/bin/activate
   
   # OR
   
   conda create --name myenv python=3.11
   conda activate myenv

2. Install dependencies:

   pip install -r requirements.txt

3. Download and organize the dataset:

   chmod +x setup.sh
   ./setup.sh

Usage

Run the meta-learning training process:

python main.py

You can modify training/evaluation settings in config.yaml.

References

1. AlexNet paper from NeurIPS Proceedings
2. Pytorch implementation of AlexNet by dansuh17
3. miniImageNet
4. miniImageNet filelists