https://github.com/angrybayblade/image_search

🔍 Image Search engine based on mnist dataset.
https://github.com/angrybayblade/image_search

autoencoder autoencoder-mnist convolutional-autoencoder image-search neural-network

Last synced: 3 months ago
JSON representation

🔍 Image Search engine based on mnist dataset.

• Host: GitHub
• URL: https://github.com/angrybayblade/image_search
• Owner: angrybayblade
• Created: 2020-05-03T12:26:43.000Z (about 5 years ago)
• Default Branch: master
• Last Pushed: 2020-05-24T11:18:36.000Z (about 5 years ago)
• Last Synced: 2025-02-08T23:28:15.999Z (5 months ago)
• Topics: autoencoder, autoencoder-mnist, convolutional-autoencoder, image-search, neural-network
• Language: Jupyter Notebook
• Homepage:
• Size: 1.12 MB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
### Image Search Using Autoencoders In MNIST Dataset 🔍

**example**

___

![](doc/search.gif)

___

Run Using

```bash

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**

    ![](doc/dense_out.png)

2. **Convolutional Autoencoder**

    ![](doc/conv_out.png)

3. **Convolutional Autoencoder With Perceptual Loss** 

    ![](doc/conv_per_out.png)

### Using different auto encoders for search

> Just open the app.py and find below code.

```python

search = DenseAutoEncoderSearch()

```

> and replace it with one of the following.

```python

search = ConvolutionalAutoEncoderSearch()

#or

search = ConvolutionalAutoEncoderWithPerceptLossSearch()

```

* UI is build in node using ReactJS⚛️