Introduction

MXNet/Gluon implementation for d-SNE: Domain Adaptation using Stochastic Neighbourhood Embedding, which was accepted at CVPR 2019 as oral presentation. d-SNE aims to perform domain adaptation by aligning the source domain and target domain in a class by class fashion. d-SNE is a supervised learning algorithm and requires a few labeled samples from the target domain for training. The semi-supervised extension can further improve its performance by incoporating unlabeled target data.

Prerequisites

Dependencies

$ pip install -r requirements.txt

# Install the correct GPU version's mxnet
$ pip install mxnet-cu100 # for CUDA 10.0

Datasets and abbreviations:

Experiments	Datasets
Digits	MNIST(MT), MNISTM(MM), SVHN(SN), and USPS(US)
Office-31	AMAZON (A), DSLR (D), and WEBCAM (W)
VisDA	Synthetic (S) and Real (R)

Due to licensing issues, we won't be able to host the datasets but provide the scripts to prepare the datasets for each experiment. Please refer to the document: how to pack dataset.

Few Shot Domain Adaptation with d-SNE

Digits Experiments: MT -> MM

Here, MNIST (MT) to MNISTM (MM) is used as an example here:

$ python tools/train.py --cfg configs/DIGITS/MT-MM.yaml

Change Logs

• [03/12/2020] Reshape the code with an experiment on MT-MM

Reference

If you find any piece of this code or the paper useful, please cite our CVPR 2019 Oral paper:

@InProceedings{Xu_2019_CVPR,
    author = {Xu, Xiang and Zhou, Xiong and Venkatesan, Ragav and Swaminathan, Gurumurthy and Majumder, Orchid},
    title = {d-SNE: Domain Adaptation Using Stochastic Neighborhood Embedding},
    booktitle = {The IEEE/CVF Conference on Computer Vision and Pattern Recognition},
    month = {June 16-20},
    year = {2019},
    pages = {2497-2506}
}