back to projects
ART-ificial
ART-ificial is an AI-powered image analysis tool that allows users to upload artwork and discover which artists' styles it most closely matches. The app uses a combination of ResNet50 feature extraction, CLIP embeddings, and Pinecone's vector database to efficiently compare uploaded images against a large Kaggle dataset of artist works. This hybrid approach produces accurate and scalable artist similarity predictions.
PythonTensorFlow/KerasPyTorchCLIPNext.jsPinecone
key features
- upload any image for fast AI-driven artist similarity matching
- CLIP-based embeddings stored and searched in Pinecone vector DB
- ResNet50 CNN feature extraction for additional visual similarity signals
- top matching artists returned with confidence scores
- clean and simple web interface
challenges & solutions
challenges
- working with a large Kaggle artist dataset and ensuring consistent preprocessing
- understanding and experimenting with embeddings for similarity search
- scaling the comparison process efficiently without slowing down the app
solutions
- used standardized preprocessing and resizing to normalize images from the Kaggle dataset
- generated CLIP embeddings for all artworks and stored them in Pinecone for fast vector search
- combined CLIP similarity results with CNN-based features for more reliable predictions