Image-Caption-Generator

About

1. The goal of this project is to create a system that generates descriptive text for a given image, essentially producing a 'caption' for the image.
2. Utilized the Flickr8k dataset, which includes 8,091 images, each accompanied by 5 captions.
3. Employed an Encoder-Decoder model architecture, using a Pre-Trained VGG16 network as the Encoder and an LSTM cell with Bahdanau Attention as the Decoder.
4. Achieved a corpus BLEU-1 score of 0.537 and BLEU-2 score of 0.315 on the test dataset(flickr8k) for caption generation accuracy.

Model Preview

Dataset

You can access the test dataset i.e., flickr8k: here.

Steps to run this on your local computer:

• Clone this repository

https://github.com/Arin13-03/Image-Caption-Generator.git

• Download the requirements of the environment using

pip install -r requirements.txt

• Download the model and place it into the same root directory as app.py. You can download the model here.
• Run app.py file to open the website

streamlit run app.py

Website Screenshots

• This is the homepage you will see in the browser. To upload your image, click on browse and select an image from your computer.
• The results will appear below your image under the 'Generated Caption' heading