Image-Caption-Generator

A WebApp that Generates Caption for Images using CNN-RNN.

Application Link: https://dev228-afk-image-caption-generator-app-c6ckdt.streamlitapp.com/

Model

This Model consists of a CNN-RNN Layer, Which is made of Keras Sequential API. it's made of the following contents:

1. CNN Encoder Model: Pretrained CNN Model, which generates Features for Input and Training Images. as an Encoder, Transfer Learning based Xception model has been used with its pretrained weights.
2. word Embedding Layer: Converts Caption into Word Embedding Tokens. it takes the input/output dimension of the Vector (32,256).
3. LSTM Decoder Model: LSTM is used as Text Sequence Processing in Encoder-Decoder Architecture, Which takes Input-pair of the feature vector of image and Partial Caption and returns Predicted Caption for input Image

Overview of the Overall Model with its Dimension is shown below:

Dataset used:

• Fliker8k (Including Images and its Text description)
• Dataset link: https://academictorrents.com/details/9dea07ba660a722ae1008c4c8afdd303b6f6e53b

Model Results:

• Some of the Captions Generated by this model are as follows:

Requirements

• Tensorflow
• Pandas
• Numpy
• Pillow
• Keras
• h5py

Usage

• Use Training_model.ipynb file for the training Model
• Use Model_Testing.ipnb file for testing model

If this Repository really helped you, please do Star to the Repo.