imgs.ai

imgs.ai is a fast, dataset-agnostic, deep visual search engine for digital art history based on neural network embeddings. It utilizes modern approximate k-NN algorithms via Spotify's Annoy library to deliver fast search results even for very large datasets in low-resource environments, and integrates the OpenAI CLIP model for text-based visual search.

Try it here on the complete Rijksmuseum and Metropolitan Museum of Art collections or sign up for an account to access more functions/datasets (institutional email address and approval required).

imgs.ai is developed by Fabian Offert, with contributions by Peter Bell and Oleg Harlamov. Get in touch at [email protected].

Local installation

Only MacOS and Linux environments are currently supported.

1. Download and install the Anaconda or Miniconda (preferred) package manager.
2. Create a Python 3.8 conda environment with conda create --yes -n imgs.ai python=3.8 and activate it with conda activate imgs.ai.
3. Clone or download the repository and run the install.sh shell script with your preferred shell. If you would like to install with GPU support, add the following parameter: cudatoolkit=10.1, where the version number is the version of your installed CUDA framework (see https://pytorch.org/ for more information).
4. To start imgs.ai, run the run.sh shell script with your preferred shell.
5. Open a web browser and navigate to localhost:5000 to see the interface.

We provide the Rijksmuseum dataset (embeddings only) for testing purposes. (Download here and extract to the models/public folder). The dataset is trained on nearly 400,000 works in the Rijksmuseum collection. This is a live dataset, images are pulled from the Rijksmuseum servers on request. Right-click an image to go to the source website on the Rijksmuseum servers.

Local training (experimental)

Please see the separate imgs.ai-custom repository to train your own dataset.