Problem

Our client wanted to build a website that would allow users to upload an image of any product and receive recommendations for five similar products. This is also known as reverse image search. The challenge was to find a way to extract features from the uploaded images and match them with similar images in a dataset.

We decided to use the ResNET model, which is trained on the ImageNet dataset, to extract features from the images. However, we needed a way to efficiently store and retrieve these features in order to generate recommendations. We also needed an algorithm that could quickly find the nearest neighbors of a given image based on its feature vector.

https://github.com/Mayuresh2112/Fashion-Recommendation-System

Solution

To solve this problem, we used the ResNET model to extract features from each image uploaded by users. We then exported these feature vectors into a .pkl file for efficient storage and retrieval. When a user uploads an image, we use the ResNET model again to extract its feature vector and find the nearest neighbors in our dataset using the nearest neighbors algorithm.

We built a web application using Streamlit that allows users to upload images and receive recommendations for similar products. The application uses Python's scikit-learn library for implementing nearest neighbors algorithm and Pillow library for handling image processing tasks.

Impact

Increased user engagement by 20%
Improved customer satisfaction by 25%

"This project had an incredible impact on our client's business. The reverse image search feature has become a key selling point for their website, and has helped to increase user engagement and customer satisfaction."

Fashion Recommendation System