GitHub Repository: dhrm1k/depict

The project introduces an AI-powered visual search tool for artworks in Indian museums, accessible on the Museums of India website. Named “Depict,” it allows users to search using natural language queries like “painting of fruit” to find visually similar pieces, leveraging OpenAI’s CLIP model for matching text descriptions to images. The initiative, inspired by Karan Janthe and executed by Dharmik, involved scraping museum data and deploying machine learning models using Python and Modal AI. Despite compute limitations restricting the dataset, users can explore a selection of artworks from six Indian museums. The system processes text input into a 512-dimensional vector to perform similarity comparisons, showcasing a practical application of vector search and embeddings. Future enhancements could include image uploading for search, sophisticated filtering, and a better mobile experience. The project is a valuable learning experience about AI’s intersection with cultural heritage.

Source link