Trendily: Fashion Recommendations

A standalone HTTP web server that can recommend similar fashion outfits, based on the the Rent the Runway inventory.

Uses multiple neural networks (with a ResNet50 backbone) behind the scenes to classify inputs by {category, texture, fabric, parts, shape}. The resulting embeddings are then used to query a pre-built nearest neighbors index for similar outputs.

Installation

Use pip to install the requirements.

Note: This repo works on Linux environment only as fastai and Annoy are NOT supported on other OS as far as I know so far.

pip install -r requirements.txt

Usage

To run the web server, simply execute flask with the main recommender app:

FLASK_APP=recommender_app flask run

flask run --host=0.0.0.0
FLASK_APP=recommender_app flask run --host=148.181.168.85

The main predictor can also be used independently of Flask, by calling get_recs:

from predict import Predict

fashion = Predict()
recs = fashion.get_recs(img_path)

Built With

• fast.ai - Deep learning library used for CNN training
• Flask - Python HTTP server
• DeepFashion - Large-scale Fashion dataset used to train the classifiers
• Annoy - Efficient Approximate Nearest Neighbors library

Files

• Outfits - Example images that can be used to test the recommendation system
• recommender_app.py - Spins up a Flask App to serve recommendations
• predict.py - Recommendation System

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

Please make sure to update tests as appropriate.

License

MIT