Skip to content

zentralwerkstatt/imgs.ai-custom

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

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 hi@imgs.ai.

This repository provides a custom training function. It is independent of the main imgs.ai repository.

Local installation (experimental)

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-custom python=3.8 and activate it with conda activate imgs.ai-custom.
  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 (GPU is strongly recommended), 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. Edit embedders.pytxt and the parameters of make_modelin run.py to fit your needs, then run run.py.

CoLab version (experimental)

Open In Colab

About

Train an imgs.ai model on your own dataset

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published