PixeLens is an AI-powered image captioning and semantic search platform. It allows users to upload images, generate natural language captions using Vision Transformers, and perform real-time text-to-image and image-to-image searches.
- Image Captioning: Generates descriptive captions for user-uploaded images via Vision Transformer models.
- Semantic Search: Supports both text-to-image and image-to-image retrieval using cosine similarity on image embeddings.
- Lightweight Storage: Stores images, captions, and embeddings in a fast, searchable SQLite database.
- Real-Time Results: Delivers search results instantly for a smooth user experience.
- Backend: Python, Gradio, Vision Transformers (ViT)
- Database: SQLite
- Search: Cosine similarity on precomputed embeddings
- Frontend: Gradio
git clone https://github.com/yourusername/pixelens.git
cd pixelens/Code
# Step 1: Create virtual environment
python -m venv venv
# Step 2: Activate it
source venv/bin/activate # On Windows: venv\Scripts\activate
# Step 3: Install requirements
pip install -r requirements.txt
# Step 4: Launch using run.sh
./run.sh
##############
1. Additional .ipynb files and scripts are kept in additional_code_files folder
2. Image_Caption_Gen_lstm_final_pipeline.ipynb is the final model code
3. augment_images.py is the script to augment images
4.images.db is a SQLITE db, it will be created automatically if not present
5. vit_caption_model.h5 is the final model weights file
6. tokenizer.p is required for the pipeline to work
pip install -r requirements.txt