wx-b/ifsl

[CVPR'22] Official PyTorch implementation of Integrative Few-Shot Learning for Classification and Segmentation

Integrative Few-Shot Learning
for Classification and Segmentation

Dahyun Kang      Minsu Cho

This repo is the official implementation of the CVPR 2022 paper: Integrative Few-Shot Learning for Classification and Segmentation.

📜 BibTex source

If you find our code or paper useful, please consider citing our paper:

@inproceedings{kang2022ifsl,
  author   = {Kang, Dahyun and Cho, Minsu},
  title    = {Integrative Few-Shot Learning for Classification and Segmentation},
  booktitle= {Proceedings of the {IEEE/CVF} Conference on Computer Vision and Pattern Recognition (CVPR)},
  year     = {2022}
}

⚙️ Conda environmnet installation

This project is built upon the following environment:

• Ubuntu 16.04
• Python 3.7
• CUDA 11.0
• PyTorch 1.7.0

The package requirements can be installed via environment.yml, which includes

• pytorch==1.7.0
• torchvision==0.8.1
• cudatoolkit==11.0.3
• pytorch-lightning==1.3.8
• einops==0.3.0

conda env create --name ifsl_pytorch1.7.0 --file environment.yml
conda activate ifsl_pytorch1.7.0

Make sure you need to replace YOURCONDAROOTDIR in the installation path with your own root dir

📚 Datasets

• PASCAL VOC 2012 and SBD
• Microsoft COCO 2014

Download the datasets by following the file structure below and set args.datapath=YOUR_DATASET_DIR:

    YOUR_DATASET_DIR/
    ├── VOC2012/
    │   ├── Annotations/
    │   ├── JPEGImages/
    │   ├── ...
    ├── COCO2014/
    │   ├── annotations/
    │   ├── train2014/
    │   ├── val2014/
    │   ├── ...
    ├── ...

We follow the dataset protocol of HSNet and PFENet.

🔍 Related repos

Our project refers to and heavily borrows some the codes from the following repos:

• [PANet]: Wang et al., Few-shot image semantic segmentation with prototype alignment, ICCV'19.
• [PFENet]: Tian et al., Prior guided feature enrichment network for few-shot segmentation, TPAMI'20.
• [HSNet]: Min et al., Hypercorrelation squeeze for few-shot segmentation, ICCV'21.

🙇 Acknowledgements

This work was supported by Samsung Advanced Institute of Technology (SAIT) and also by Center for Applied Research in Artificial Intelligence (CARAI) grant funded by DAPA and ADD (UD190031RD). We also thank Ahyun Seo and Deunsol Jung for their helpful discussion.