Fashion-Gen: The Generative Fashion Dataset and Challenge by Negar Rostamzadeh, Seyedarian Hosseini, Thomas Boquet, Wojciech Stokowiec, Ying Zhang, Christian Jauvin, Chris Pal.
Forked from https://github.com/taoxugit/AttnGAN
Pytorch implementation for reproducing AttnGAN results in the paper AttnGAN: Fine-Grained Text to Image Generation with Attentional Generative Adversarial Networks by Tao Xu, Pengchuan Zhang, Qiuyuan Huang, Han Zhang, Zhe Gan, Xiaolei Huang, Xiaodong He. (This work was performed when Tao was an intern with Microsoft Research).
python 2.7
Pytorch
In addition, please add the project folder to PYTHONPATH and pip install the following packages:
python-dateutileasydictpandastorchfilenltkscikit-image
Download the dataset images and extract them to data/fashiongen/.
fashiongen_256_256_train.h5
fashiongen_256_256_validation.h5
fashiongen_consume_data_example.pdf
-
Pre-train DAMSM models:
python pretrain_DAMSM.py --cfg cfg/DAMSM/fashiongen2.yml --gpu 0 -
Train AttnGAN models:
python main.py --cfg cfg/fashiongen2_attn2.yml --gpu 0
fashiongen2.yml config file will train only on 7 categories (representing 61% of the dataset). To train on the complete dataset use fashiongen.yml.
Training with fashiongen2.yml on a single GPU NVidia GTX1070 Ti takes 8 hours per epoch with the default settings.
Below, a generated image after 51 epochs of training with fashiongen2.yml config file. More epochs of training would be required to get more realistic results.
In addition to the two config files in the cfg folder, you will find code specific to the fashion-gen dataset in the code folder. The main file is dataset_fashiongen2.py and the various Jupyther Notebook files to explore the dataset.