Compositional Prototypical Networks for Few-Shot Classification

For details see the paper (The paper has been accepted by AAAI 2023)

Official PyTorch implementation of Compositional Prototypical Networks for Few-Shot Classification

0. Overview

1. Requirements

• python 3.8.13
• pytorch 1.8.0

2. Install

Clone this repository:

git clone https://github.com/fikry102/CPN.git
cd CPN

3. Datasets

Datasets can be downloaded using CUB and SUN GoogleDrive

We recommend the following structure to place the datasets:

---

|--data

​ ||--cub

​ ||--sun

|--CPN

---

4.Run

cd scripts 

#CUB, pre-training
python 01_CUB_stage1.py

#CUB, 5way 1shot, meta-training and meta-testing 
python 01_CUB_stage2_5W1S.py

#CUB, 5way 5shot, meta-training and meta-testing
python 01_CUB_stage2_5W5S.py

or

sh run_cub.sh

5. Models

We provide off-the-shelf models in Models GoogleDrive

Acknowledgement

This repo is based on SEGA, Dynamic Few-Shot Visual Learning without Forgetting and MetaOptNet.

Thanks for their wonderful works.

Contact

If you have any question about this project, please feel free to contact [email protected].