🖼️ Image Captioning with BLIP (Vision-Language Model)

This project demonstrates how to generate captions for images using the BLIP (Bootstrapped Language-Image Pretraining) model by Salesforce, powered by the 🤗 Hugging Face Transformers library.

It is designed to run in Google Colab and uses a dataset of images (such as a subset of Flickr8k) to generate natural language captions.

---

📌 Features

• ✅ Uses Salesforce/blip-image-captioning-base for image captioning
• ✅ Automatically loads and processes images from a ZIP file
• ✅ GPU-accelerated via Google Colab
• ✅ Shows sample outputs using matplotlib
• ✅ Clean and modular Python code

---

📁 Dataset

The dataset used is a 2,000-image subset of the Flickr8k dataset.

📥 Download here:
https://www.kaggle.com/datasets/sanjeetbeniwal/flicker8k-2k

Expected structure inside the ZIP file:

Flickr8k\_2k.zip
└── Flicker8k\_2kDataset/
├── image1.jpg
├── image2.jpg
└── ...

Upload this ZIP file to your Colab environment before running the notebook.

---

🛠️ Dependencies

The following Python packages are required:

pip install torch torchvision torchaudio
pip install transformers
pip install matplotlib

All dependencies are automatically installed in the Colab notebook.

---

🚀 How It Works

1. Setup: Install required libraries and enable GPU runtime.
2. Dataset Unzipping: Upload and extract the dataset in Colab.
3. Model Loading: Load BLIP processor and model to GPU.
4. Captioning: Select and caption random images.
5. Visualization: Display images with generated captions using matplotlib.

---

📸 Sample Output

Below is an example of the model generating a caption for an image from the dataset:

Image: screenshot_20250721_235853.jpg Generated Caption: `a child sitting in a play area'

---

💡 Model Info

• Model: Salesforce/blip-image-captioning-base
• Library: Hugging Face Transformers
• Pretrained for general image-to-text tasks.

---

▶️ Usage Instructions

1. Open the notebook in Google Colab.

2. Upload your dataset ZIP file to Colab (Flickr8k_2k.zip).

3. Set runtime to GPU:

   • Runtime → Change runtime type → GPU

4. Run all cells sequentially.

5. View the images and their generated captions.

---

📄 License

This project is for educational and research purposes. It uses publicly available pretrained models under their respective licenses.