Generalizable Two-Branch Framework for Image Class Incremental Learning （ICASSP'24）

This repository is Pytorch code for our proposed Generalizable Two-Branch Framework for Image Class Incremental Learning (G2B).

Paper link: https://arxiv.org/abs/2402.18086

1.Environment Setup

The code and models were tested on Linux Platform with two GPU (RTX3080Ti). First creating a conda environment with all the required packages by using the following command.

conda env create -f environment.yml

It creates a conda environment named G2B. You can activate the conda environment with the command:

conda activate G2B

In the following sections, we assume that this conda environment is in use.

Potential Compatibility Issues: 1.If you see the following error, it usually mean the PyTorch package incompatible with the infrastructure.

RuntimeError: CUDA error: no kernel image is available for execution on the device

For example, your machine supports CUDA == 11.1, install a PyTorch package using CUDA11.1 instead:

pip uninstall torch
pip install torch==1.8.1+cu111 torchvision==0.9.1+cu111 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html

2.Run

Training and testing of the proposed method are reproduced on CIFAR100 10-task class incremental learning (CIL):

G2B(DER)

# Model Training
cd ./CNN
python main.py --config=./exps/cifar100/G2B_DER.json

G2B(DyTox)

# Model Training
cd ./VIT
bash train.sh 0,1 \
  --options options/data/cifar100_10-10.yaml options/data/cifar100_order1.yaml options/model/cifar_dytox.yaml \
  --name G2B_dytox \
  --data-path ./data \
  --output-basedir ./checkpoints \
  --memory-size 2000 \
  --add_mask

3.Results

The reproduced results of CIFAR100 10-task CIL:

Task	1	2	3	4	5	6	7	8	9	10	Avg.
G2B(DER)	94.6	87.75	82.23	77.88	76.8	75.55	74.68	73.24	71.03	68.98	78.26
G2B(DyTox)	90.9	88.25	83.67	79.22	77.74	71.3	69.17	65.45	63.49	62.04	75.12

Note: Different pytorch versions may lead to slightly different results. (pytorch ver. >= 1.8.1 required).

Citation

If you find this code useful, please kindly cite the following paper:

@article{wu2024general,
  title={Generalizable Two-Branch Framework for Image Class Incremental Learning},
  author={Wu, Chao and Chang, Xiaobin and Wang, Ruixuan},
  booktitle = {IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP)},
  year={2024}
}

ftd-Wuchao / G2B