• We integrate (DVBPR + VGG16)/GAN model to generate the best outfit according to user preference.

The code is tested under a Linux AWS P2 with GPU power.

Requirements:

• TensorFlow 1.3
• Numpy
• PIL

The four fashion datasets:

• AmazonFashion (3.3GB) : 64K users, 234K images, 0.5M actions
• AmazonWomen (6.2GB): 97K users, 347K images, 0.8M actions
• AmazonMen (2.1GB): 34K users, 110K images, 0.2M actions
• Tradesy (3.4GB): 33K users, 326K images, 0.6M actions

Step 1: Train DVBPR:

cd DVBPR
python main.py

The default hyper-parameters are defined in main.py, you can change them accordingly. AUC (on validation and test set) is recorded in DVBPR.log.

Step 2: Train GANs:

cd GAN
python main.py --train True

The default hyper-parameters are defined in main.py, you can change them accordingly. Without '--train True', it will load a trained model and generated images for each category (stroed in folder samples).

Step 3: Preference Maximization:

cd PM
python main.py

PM is based on pretrained DVBPR and GAN models. It will randomly pick a user for each category, and show the generated images through the optimization process.