pip install -r requirements.txt

conda env create -f environment. yml

source activate SSDA

The code is written for Pytorch 0.4.0, but should work for other version with some modifications.

To get data, run

sh download_data.sh

The images will be stored in the following way.

./data/multi/real/category_name,

./data/multi/sketch/category_name

The dataset split files are stored as follows,

./data/txt/multi/labeled_source_images_real.txt,

./data/txt/multi/unlabeled_target_images_sketch_3.txt,

./data/txt/multi/validation_target_images_sketch_3.txt.

With regard to office and office home dataset, store the image files in the following ways,

./data/office/amazon/category_name, ./data/office_home/Real/category_name,

We provide the split of office and office-home.

To run training using alexnet,

sh run_train.sh gpu_id method alexnet

where, gpu_id = 0,1,2,3...., method=[MME,ENT,S+T].

This repository is build on top of MME repo by Kuniaki Saito and Donghyun Kim.