🔍 Image Search with CLIP (Contrastive Language–Image Pretraining)

This project allows you to search images using natural language queries by leveraging the power of OpenAI's CLIP model. It loads a folder of images, generates embeddings, and then finds the most similar images to your text description.

Built for use in Google Colab or local Python environments.

---

🧠 Powered by CLIP

CLIP (Contrastive Language–Image Pretraining) is a powerful vision-language model that understands both images and text. This project uses CLIP to extract image embeddings and compare them with text queries using cosine similarity.

---

🗂️ Project Workflow

1. Load CLIP Model
   Loads openai/clip-vit-base-patch32 via 🤗 Hugging Face.

2. Load Images
   Scans the images/ directory and loads .jpg files.

3. Extract Embeddings
   Images are processed and converted into feature vectors.

4. Search
   User types a natural language query. Top-3 most similar images are returned with similarity scores.

---

🖼️ Sample Output

Loading CLIP model and processor...
CLIP model loaded successfully!

Loading images from 'images'...
Loaded: 310715139\_7f05468042.jpg
Loaded: 311619377\_2ba3b36675.jpg
...

Extracting image features (embeddings)... This might take a moment.
Image features extracted.

Enter your search query (e.g., 'a cat in a garden', 'a fast car', or 'exit' to quit): image with black dog

Top 3 results for 'image with black dog':

* images/333973142\_abcd151002.jpg (Similarity: 0.2749)
* images/326456451\_effadbbe49.jpg (Similarity: 0.2684)
* images/3101796900\_59c15e0edc.jpg (Similarity: 0.2520)

---

🔧 Requirements

Install dependencies with:

pip install torch torchvision torchaudio
pip install transformers
pip install pillow

---

📁 Folder Structure

project/
│
├── main.py               # Main search script
├── images/               # Folder containing all images
│   ├── image1.jpg
│   ├── image2.jpg
│   └── ...
└── README.md

---

▶️ How to Use

1. Clone the repo or upload your images to a folder named images/.
2. Run the script and wait for embeddings to be generated.
3. Type any text prompt (like "a dog on grass") and view the results.
4. Type 'exit' to quit the loop.

---

💬 Example Queries

• "a man riding a bicycle"
• "a dog laying on the ground"
• "a child playing with a toy"
• "a group of people walking"

---

📄 License

This project is for educational and research purposes only. CLIP is provided under the license of OpenAI and Hugging Face.

---