Deep-Learning-based-Image-Caption-Generator

🖼️ Image Caption Generator (CNN + LSTM)

This project implements an Image Caption Generator using Deep Learning techniques.
It combines a Convolutional Neural Network (CNN) for image feature extraction with a Recurrent Neural Network (LSTM) for generating descriptive captions in natural language.

---

Project Overview

• Dataset: Flickr8k (8,000 images with 5 captions each)
• Model: CNN (InceptionV3) + LSTM
• Frameworks: Python, TensorFlow/Keras
• Goal: Generate accurate captions for unseen images by learning the mapping between image features and natural language.

---

Features

• Image preprocessing and feature extraction using pre-trained InceptionV3
• Caption preprocessing with tokenization and padding
• Sequence modeling with LSTM
• Evaluation using BLEU scores
• Inference script to generate captions for custom images

---

Folder Structure

data/ # Dataset: images and captions src/ # Scripts: utils, feature extraction, model, training, inference notebooks/ # Interactive notebooks for exploration, training, inference requirements.txt # Python dependencies

---

Dataset

• Images: 8,000 images of everyday scenes.
• Captions: 5 captions per image (Flickr8k.token.txt).

Instructions:

1. Download Flickr8k dataset.
2. Place images in data/Images/ and captions in data/Flickr8k.token.txt.

---

Setup

1. Clone the repo: git clone <your_repo_link> cd Image-Caption-Generator

2. Install dependencies:

   Training

• Run notebooks/03_model_training.ipynb to train the model.
• Training and validation loss will be plotted automatically.
• Model weights will be saved to models/model_weights.h5.

---

Inference / Demo

• Run notebooks/04_inference_demo.ipynb to generate captions for sample images.
• Visualize results inline or save them in examples/.

---

References

• Flickr8k Dataset