Genj1Kai/ActiveMatch

Code for the paper: "ActiveMatch: End-to-end Semi-supervised Active Representation Learning"

ActiveMatch

Code for the paper: "ActiveMatch: End-to-end Semi-supervised Active Representation Learning" by Xinkai Yuan, Zilinghan Li, and Gaoang Wang.

Installation

conda env create --name activematch --file environment.yml
conda activate activematch
python train.py

Running

Example

Train our model with 40 to 200 labels of CIFAR-10:

python train.py --dataset cifar10 --num-labeled 40 --batch-size 64 --lr 0.03 --expand-labels --seed 5 --out results/cifar10@40-200 --stop-active 200 --num-sample 32 --epoch-warmup 15

Train our model with 200 to 1000 labels of CIFAR-100:

python train.py --dataset cifar100 --num-labeled 200 --batch-size 64 --lr 0.03 --expand-labels --seed 5 --out results/cifar10@200-1000 --stop-active 1000 --num-sample 4 --epoch-warmup 15

Citing this work

@INPROCEEDINGS{9898008,
  author={Yuan, Xinkai and Li, Zilinghan and Wang, Gaoang},
  booktitle={2022 IEEE International Conference on Image Processing (ICIP)}, 
  title={ActiveMatch: End-To-End Semi-Supervised Active Representation Learning}, 
  year={2022},
  volume={},
  number={},
  pages={1136-1140},
  doi={10.1109/ICIP46576.2022.9898008}
}