Image search with Deep Learning

Ever though how Google's Image Reverse Search or Pinterest's Visual Search algorithms work? Well, this project is one way to build such a system.

Your search image first goes through a Convolutional Autoencoder. What comes out is a small vector representation of the image which captures all the relevant features in it. This little content vector is searched against a previously indexed image database using Approximate Nearest Neighbor algorithm. The results are returned and ranked in the UI as shown below:

(Note: Examples show indexed images of smaller resolutions)

Searching for planes:

Searching for trucks: