sahu1shivani / Image-captioning-VGG16

Trained code on COCO dataset with attention model for Image Captioning.Usage of pre-trained model like VGG-16

Image-captioning-VGG16

This repository contains the implementation of an image captioning system using a deep learning model that combines a Convolutional Neural Network (CNN) encoder and a Transformer-based decoder.For our image based model (viz encoder) – we usually rely on a Convolutional Neural Network model. And for our language based model (viz decoder) – we rely on a Recurrent Neural Network

Overview

The project uses a CNN encoder (based on InceptionV3) to extract features from images, and a Transformer-based decoder to generate captions. The model is trained on a subset of the MS COCO dataset.

Getting Started

Prerequisites

• Python 3.8 or later
• TensorFlow 2.4 or later
• NumPy
• Pandas
• Matplotlib
• Pillow

Model Architecture

The model architecture consists of:

• CNN Encoder: A modified InceptionV3 model that extracts features from images.
• Transformer Encoder Layer: Applies layer normalization, dense layers, and multi-head attention to the encoder input.
• Embeddings: Combines token embeddings with position embeddings.
• Transformer Decoder Layer: Includes multi-head attention layers (for self-attention and encoder-decoder attention), layer normalization, and feed-forward networks.
• ImageCaptioningModel: The main model that combines the CNN encoder, transformer encoder, and decoder.

Acknowledgments

• The MS COCO dataset for providing the images and captions for training the model.
• The TensorFlow team for providing an excellent deep learning framework.