Personalized Photo Recommendation

This repository includes work for the Aesthetic Features for Personalized Photo Recommendation.

Data

You need to save your data in the folder data/. If you only have one dataset, you can save under the folder with name 'sample/'. If you have multiple datasets, you can save multiple them in separate folders with any name format.

Under each folder (for one datset), you have a folder named 'validation/' for parameter tuning, 'test/' for evaluation, and photos to save all photos to be used. Under each of the folder, there will be matrix_train.npz for training and matrix_test.npz for testing.
The photo data should be all in .jpg format.

Create Aesthetic Feature Embedding

Create Color Embedding

You can create color embedding for all your photos with the script below:

cd aesthetic_features
python create_color_histogram.py --data-folder data/

By default, the program will create RGB, HSV, and HLS embedding respectively for all photos.

Create Style Embedding

You can run style embedding for all your photos with the script below:

cd aesthetic_features
python create_style_embedding.py --data-folder data/

Evaluate Models

You can evaluate with all the models on your data with the script below:

python model_evaluation.py --data-folder data/ --task test