Skip to content

Latest commit

 

History

History
12 lines (9 loc) · 963 Bytes

File metadata and controls

12 lines (9 loc) · 963 Bytes

Product-Recommendation-Engine-Based-on-Visual-Similarity

The subject of this repository was to develop a very basic recommender engine for fashion products. For this case, the system should recommend goods that look similar. The idea behind is very simple: if a customer is showing interest towards a specific product by browsing its page, he may also be interested by products that are visually alike. We will implement CNNs with an image feature extraction approach, with Keras and TensorFlow packages. The recommender engines were based on pre-trained models: ResNet50 and VGG16.

image

The recommender systems are able to find similar products accurately: most of the time the retrieved products have the same purpose and look very similar. They certainly work well in terms of shapes; patterns are not always recognized.