Task-Aware Variational Adversarial Active Learning - Official Pytorch implementation of the CVPR 2021 paper

Kwanyoung Kim, Dongwon Park, Kwang In Kim, Se Young Chun

Official Pytorch implementation for the paper presented on CVPR 2021 titled "Task-Aware Variational Adversarial Active Learning".

Abstract

We propose task-aware variational adversarial AL (TA-VAAL) that modifies task-agnostic VAAL, that considered data distribution of both label and unlabeled pools, by relaxing task learning loss prediction to ranking loss prediction and by using ranking conditional generative adversarial network to embed normalized ranking loss information on VAAL. Our proposed TA-VAAL outperforms state-of-the-arts on various benchmark datasets for classifications with balanced / imbalanced labels as well as semantic segmentation and its task-aware and task-agnostic AL properties were confirmed with our in-depth analyses.

Prerequisites:

• Linux or macOS
• Python 3.5/3.6
• CPU compatible but NVIDIA GPU + CUDA CuDNN is highly recommended.
• pytorch 0.4.1
• cuda 8.0
• Anaconda3

Requirements

To install virtual enviornment for requirements:

conda env create -f TAVAAL.yaml

📋 if you already conda, you can activate virtual experiment settings

To activate virtual enviornment:

conda activate TAVAAL

Running code

To train the model(s) and evaluate in the paper, run this command:

python main.py -m TA-VAAL -d cifar10 -c 5 # Other available datasets cifar100, fashionmnist, svhn

if you want to experiment about cifar_imbal. you can run this command:

python main.py  -d cifar10 cifar10im