DANN-PyTorch 🔥

PyTorch implementation of DANN (Domain-Adversarial Training of Neural Networks)

Unsupervised Domain Adaptation by Backpropagation
Yaroslav Ganin, Victor Lempitsky
In PMLR-2015

Domain-Adversarial Training of Neural Networks
Yaroslav Ganin et al.
In JMLR-2016

Getting started

Installation

Install library versions that are compatible with your environment.

git clone https://github.com/NaJaeMin92/pytorch-DANN.git
cd pytorch-DANN
conda create -n dann python=3.7
conda activate dann
pip install -r requirements.txt

Recommended configuration

python=3.7
pytorch=1.12.1
matplotlib=3.2.2
sklearn=1.0.2

Usages

Running the code below will execute both source-only and DANN training and testing:

python main.py
# You can adjust training settings in 'params.py', including batch size and the number of training epochs.

t-SNE (t-distributed Stochastic Neighbor Embedding)

Our code includes the functionality to visualize t-SNE, both before and after the process of domain adaptation using sklearn.manifold.

Experimental results

MNIST -> MNIST-M

Method	Test #1	Test #2	Test #3	Test #4	Test #5	Avg.
Source Accuracy	89	98	98	90	98	61.2
Target Accuracy	47	56	54	46	53	51.2

DANN

Method	Test #1	Test #2	Test #3	Test #4	Test #5	Avg.
Source Accuracy	96	96	97	97	96	96.4
Target Accuracy	83	78	80	80	78	79.8
Domain Accuracy	60	60	61	64	61	61.2