Disentanglement based Active Learning.
Adarsh K, Silpa V S, S Sumitra
IJCNN 2021
- Ubuntu
- Python 3
- NVIDIA GPU + CUDA CuDNN
Disentanglement based Active Learning (DAL), a new active learning technique based on self-supervision which leverages the concept of disentanglement. Instead of requesting labels from human oracle, our method automatically labels majority of the datapoints, thus drastically reducing the human labeling budget in GAN based active learning approaches
- Clone this repo:
git clone https://github.com/kadarsh22/disentanglement_based_active_learning.git
cd disentanglement_based_active_learning-
Install dependencies:
- Install dependcies to a new virtual environment.
pip install -r requirements.txt
-
Download resources:
- Download pretrained models from [here] (https://drive.google.com/open?id=1M_5ZumHrNjn-_rTiBA6nnsHiYi9p-TbU)
- Place the pretrained models from here utils/fashionmnist/trainedmoddels
To run DAL :
python main.py --dataset 'mnist' --gan_type 'infoGAN' --output_activation 'sigmoid' --data_size 10000
python main.py --dataset 'fashion-mnist' --gan_type 'infoGAN' -output_activation 'tanh' --data_size 10000