Skip to content

prasadgujar/CapSearch

Repository files navigation

CapSearch

An Automated Image Caption Generation and Caption Based Image Retrieval Application.

contents

Introduction

This is a python (Flask Application) based Automated Image Caption and Image Retrieval model which makes use of deep learning image caption generator. It uses a merge model comprising of Convolutional Neural Network (CNN) and a Long Short Term Memory Network (LSTM) . The dataset used here is Flickr8K dataset.

Features

This model can be used via GUI. In model-

  • Automated Caption Generation (Offline) - Upload Image and retrive automated caption based on image features.
  • Caption Based Image Search (Similar Images) - Given Text Based Query and it will return similar images based on image caption and similarity.

Video

CapSearch

Dataset and Pre-trained models

Packages Required:

  • Anaconda
  • Keras with Tensorflow Backend (Python 3.6)
  • Flask

What you can expect in future versions?

  • Make a highly scalable REST API which accepts the image and returns the caption of the image
  • Make a dashboard through which the training of the captioner could be done on custom datasets.
  • Introduce unit tests and logging to enable smooth debugging.
  • Improve the caption based image search part for the more accuracy.
  • Make a dashboard through which user can manage their image database.
  • Improve the UI part of the application.
  • Change the architecture of image captioner in order reduce the memory footprint required by the current pre trained models
  • Further development may also include working on improvising with more accurate predictions and search results

Paper Publication:

Contributor

Contribute

  • Fork this repository and contribute.
  • Feel free to report bugs.
  • All types of feedbacks are welcome

Acknowledgement

  • A special thanks to Machine Learning Mastery without which we couldn't have thought about the right approach to tackle this problem.