Image Search Using Autoencoders In MNIST Dataset 🔍

example

---

---

Run Using

python3 app.py

Then open http://localhost:8080 on your browser.

Search engine architecture uses a auto encoder to encode images into a 128-D vectors and clusters them using KMeans.

When querying an image image is fed into the autoencoder which returns a 128-D vector, the vector is used to find which cluster the image belongs to and then some of the images from the cluster are fetched as result.

Auto-Encoders Used In Search Engine

1. Dense Autoencoder

   

2. Convolutional Autoencoder

   

3. Convolutional Autoencoder With Perceptual Loss

   

Using different auto encoders for search

Just open the app.py and find below code.

search = DenseAutoEncoderSearch()

and replace it with one of the following.

search = ConvolutionalAutoEncoderSearch()
#or
search = ConvolutionalAutoEncoderWithPerceptLossSearch()

• UI is build in node using ReactJS⚛️