HUTTON9453/Active-DA

Improving Clustering Uncertainty-weighted Embeddings for Active Domain Adaptation

Overview

Active-DA is a Python package designed to make active domain adaptation easier for researchers and real-world users. Active-DA based on DeepAL. In our work, we analyze the state-of-the-art of active domain adaptation, ADA-CLUE. We observed that uncertainty is not reliable in early stage of training. Thus in this package, we proposed a intuition method, Loop threshold, to improve the performace of CLUE. We introduce a threshold to control when to using uncertainty-weighted clustering or constant-weighted clustering. We called Clustering Uncertainty-weighted Embeddings with Loop Threshold as simple solution. And in this package we not only implement popular active learning methods for domain adaptation but also provides benchmark results on cross-domain dataset.

Requirements

• Python 3.7
• Pytorch 1.1
• PyYAML 5.1.1

Strategy

We provide some baseline strategies as well as some state-of-the-are strategies in this package as the following:

• Random Sampling
• Margin Sampling
• Entropy Sampling
• Coreset
• KMeans Sampling
• BADGE
• AADA
• CLUE
• Density Weighted Uncertainty Sampling (the variant of CLUE)(Link coming soon)
• CLUE with Loop threshold(Link coming soon)

Dataset

The structure of the dataset should be like

Office-31
|_ category.txt
|_ amazon
|  |_ back_pack
|     |_ <im-1-name>.jpg
|     |_ ...
|     |_ <im-N-name>.jpg
|  |_ bike
|     |_ <im-1-name>.jpg
|     |_ ...
|     |_ <im-N-name>.jpg
|  |_ ...
|_ dslr
|  |_ back_pack
|     |_ <im-1-name>.jpg
|     |_ ...
|     |_ <im-N-name>.jpg
|  |_ bike
|     |_ <im-1-name>.jpg
|     |_ ...
|     |_ <im-N-name>.jpg
|  |_ ...
|_ ...

The "category.txt" contains the names of all the categories, which is like

back_pack
bike
bike_helmet
...

Training

./experiments/scripts/train.sh ${config_yaml} ${gpu_ids} ${domain_adaptation_solver} ${dataset} ${active_learning_method} ${experiment_name} ${seed}

For example, for the Office-31 dataset,

./experiments/scripts/train.sh ./experiments/config/Office-31/MME/office31_train_amazon2dslr_cfg.yaml 0 CAN office31 OurSampling office31_a2d 1126

The experiment log file and the saved checkpoints will be stored at ./experiments/ckpt/${experiment_name}

Contact

If you have any questions, please contact me via r08922134@csie.ntu.edu.tw.

Acknowledgement

The authors thank members of the Computational Learning Lab at National Taiwan University.