We have created a system for one online clothing retailer in middle east to show customers similar dresses based on their sent photos and styles.

We have created a model that is capable of doing large-scale visual recommendation which is one that can incorporate visual signals directly into the recommendation objective. Essentially, a user interested in buying a particular item from the screen may want to explore visually similar items before finishing her purchase. These could be items with similar colors, patterns, and shapes.

This system uses deep learning and specifically ResNet deep neural network architecture. We did attribute prediction of clothing by using DeepFashion dataset and Tensorflow Version Two’s Keras implementation to pull off this task and we could identify similar dresses for people with almost %90 accuracy and user approval which was a huge success for such a system.

We deployed the system to a machine learning pipeline on both Amazon AWS as the primary back-end and Google GCP as the backup cloud provider.

The system also has a tuning mechanism of voting by trusted circle of moderators and power users to be trained on the go and show case better results throughout the utilization period.