📸 Image Caption Generator

A Generative AI project that automatically describes the content of an image using Deep Learning. It combines Computer Vision (InceptionV3) and Natural Language Processing (LSTM) to generate accurate, human-like captions.

🚀 Live Demo

The model is deployed and running live! You can test it with your own images here: 👉 Click here to try the Live App on Hugging Face

---

🧠 Technical Architecture

This project uses an Encoder-Decoder architecture:

1. Image Encoder (InceptionV3):
   • We use a pre-trained InceptionV3 model (trained on ImageNet) to extract high-level visual features from images.
   • The last classification layer is removed, leaving us with a feature vector of shape (2048,).
2. Sequence Decoder (LSTM):
   • The extracted image features are passed to an LSTM (Long Short-Term Memory) network.
   • The LSTM learns to generate a sequence of words (caption) based on the image features and the previous words generated.

Model Pipeline: Input Image ➡️ InceptionV3 ➡️ Feature Vector ➡️ LSTM ➡️ Predicted Caption

---

📂 Dataset

The model was trained on the Flickr8k Dataset, which consists of:

• 8,000 images (6,000 training, 1,000 val, 1,000 test).
• 5 captions per image (Total 40,000 captions).

> Note: Due to size constraints, the raw dataset is not included in this repository. You can download it from Kaggle and place it in the src/ folder.

---

🛠️ Installation & Setup

To run this project locally on your machine:

1. Clone the repository:

   git clone https://github.com/Marshal-GG/Advanced-Image-Captioning-System.git
   cd Advanced-Image-Captioning-System

2. Install dependencies:

   pip install -r requirements.txt

3. Download the Data:

   • Download Flickr8k images and Flickr8k.token.txt.
   • Place them in the src/ folder (or update paths in the notebook).

4. Run the Training Notebook:

   • Open main.ipynb to see the data preprocessing, model training, and evaluation steps.

5. Run the App:

   python app.py

---

📊 Results

• Metric: The model effectiveness is evaluated using qualitative analysis (visual inspection).
• Sample Output:
  • Input: Image of a 2 dogs running on grass.
  • 
  • Output: "Two dogs are playing together in the grass"

---

🤝 Connect

If you have any questions about this project or want to discuss Generative AI, feel free to connect!

• LinkedIn: [https://www.linkedin.com/in/rupam-g/]
• Email: [marshalgcom@gmail.com]