Skip to content

This repo contains the ml experiment that was used to find all the similar images in the database.

Notifications You must be signed in to change notification settings

Aakash-ML/SimilarImageSearch

Repository files navigation

SimilarImageSearch

This repo contains the ml experiment that was used to find all the similar images in the database.

Please Read PDF document in the repo

Code Details:-

Task 1

To find similar images.

All the .ipynb are named as:-

  • Avantari_TASK1_v1: Best performance, Used pretrained MobileNet, Imagesize 224x224x3

  • Avatari_TASK1_v2: Use of simple autoencoder, not the best

  • Avatari_TASK1_v2_3: Use of a bigger autoencoder without dropout layer, reconstruction better than previous version but performance for similarity check still not comparable to pre-trained model.

  • Avantari_AutoEncoder_V0.1: Have tried to create a better performing Autoencoder, And it thus perform better. Please have a look and comment.

Task 2

Image clustering

  • Avantari_TASK2_V1: Used pretrained MobileNet to extract feature then DBSCAN to cluster images.

Reference

https://towardsdatascience.com/building-a-similar-images-finder-without-any-training-f69c0db900b5

About

This repo contains the ml experiment that was used to find all the similar images in the database.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published