Home

Awesome

Pytorch Implementation for FBL

[CVPR-2023] Federated Incremental Semantic Segmentation

This is the official implementation code of our paper "Federated Incremental Semantic Segmentation" accepted by CVPR-2023.

You can also find the arXiv version with supplementary material at here.

This repository contains all of our source code. We sincerely thank the help of other repositories: Cermelli et al.'s repository, Douillard et al.'s repository, and Zhang et al.'s repository.

Brief Introduction

overview

Requirements

Recommend: You can directly use our encapsulated files (FCIL.zip) to create an environment. The steps are as follows:

1) unzip FCIL.zip
2) conda create -n FCIL --clone ./FCIL
3) pip uninstall matplotlib
4) pip install matplotlib==3.3.4

Dataset

Two scripts are available to download ADE20k and Pascal-VOC 2012, please see in the data folder.

How to perform training

The default is to use a pretraining for the backbone used, that is searched in the pretrained folder of the project. We used the pretrained model released by the authors of In-place ABN, that can be found here: link(place the downloaded "resnet101_iabn_sync.pth.tar" in the "pretrained" folder of the project).

We also provided 4 trained base models corresponding to the 4 settings used in our paper (i.e., VOC 4-4/8-2/10-1, ade 100-10)(place the downloaded base models in the "checkpoints/seed_2023-ov" folder of the project). Your future development can be directly based on our trained base models, just add "--base_weights" option to your own shell scripts.

Example commands

nohup bash scripts/voc/ours_4-4_overlap.sh >logs/voc/ours_4-4_overlap.log 2>&1 &
nohup bash scripts/ade/ours_100-10_overlap.sh >logs/ade/ours_100-10_overlap.log 2>&1 &

Our Related Works

Some related works about this topic are listed as follows:

  1. [CVPR-2022] Federated Class-Incremental Learning [Code]

  2. [TPAMI-2023] No One Left Behind: Real-World Federated Class-Incremental Learning [Code]

Citations:

If you find this code is useful to your research, please consider to cite our papers.

@InProceedings{dong2023federated_FISS,
    author = {Dong, Jiahua and Zhang, Duzhen and Cong, Yang and Cong, Wei and Ding, Henghui and Dai, Dengxin},
    title = {Federated Incremental Semantic Segmentation},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month = {June},
    year = {2023},
    pages = {3934-3943}
}
@InProceedings{dong2022federated_FCIL,
    author = {Dong, Jiahua and Wang, Lixu and Fang, Zhen and Sun, Gan and Xu, Shichao and Wang, Xiao and Zhu, Qi},
    title = {Federated Class-Incremental Learning},
    booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month = {June},
    year = {2022},
}
@ARTICLE{9616392,
  author={Dong, Jiahua and Li, Hongliu and Cong, Yang and Sun, Gan and Zhang, Yulun and Van Gool, Luc},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)}, 
  title={No One Left Behind: Real-World Federated Class-Incremental Learning}, 
  year={2023}
}

Contact: