SemArt Project

This project, developed for INFM2100 - Aktuelle Themen der Softwareentwicklung, is a visual art similarity search application. It leverages deep learning models and vector embeddings to identify and retrieve visually similar artworks from a dataset. The code relies on Milvus for storing and querying embeddings and ResNet-50 as a feature extractor model.

Project Description

The SemArt Project enables the analysis and search for artworks with similar visual attributes. By using pre-trained neural networks (ResNet-50) for feature extraction and Milvus for vector similarity search, it can recommend artworks based on visual features, type, and school (e.g., religious, portrait).

Key Features

• Image Preprocessing: Uses opencv and PIL for image processing.
• Embeddings: Creates embeddings using ResNet-50 from the transformers library.
• Vector Database: Manages embeddings in Milvus, a high-performance vector database.
• Similarity Search: Searches similar images based on predefined type and school categories.

Requirements

• Python 3.9+
• Libraries:
  • transformers: for feature extraction
  • numpy, pandas: data handling and analysis
  • Milvus: vector database for embedding storage and similarity search
  • plotly, matplotlib: visualization tools for data analysis
  • opencv, PIL: image processing