iNaturalist keeps full species classification models private

Summary

computervision We're doing some computer vision stuff at iNat. models iNaturalist makes a subset of its machine learning models publicly available while keeping full species classification models private due to intellectual property considerations and organizational policy. We provide “small” models trained on approximately 500 taxa, including taxonomy files and a geographic model, which are suitable for on-device testing and other applications. Additionally, researchers have independently developed and released open-source models based on iNaturalist data, which can be found in various model distribution venues (for example Hugging Face or Kaggle). os x dependencies brew install libmagic python python3 -m venv venv source ./venv/bin/activate pip3 install -U pip pip3 install -r requirements.txt installation Here's a rough script for OS X assuming you already have homebrew, Python, and virtualenv installed. # Get dependencies brew install libmagic # Get the repo git clone git@github.com:inaturalist/inatVisionAPI.git cd inatVisionAPI/ # Set up your python environment python3 -m venv venv source venv/bin/activate pip3 install -U pip pip3 install -r requirements.txt # Copy your config file (and edit, of course) cp config.yml.example config.yml # Run the app python app.py Now you should be able to test at http://localhost:6006 via the browser. Notes If the device you're installing on has AVX extensions (check flags in /proc/cpuinfo), try compiling tensorflow for better performance: https://www.tensorflow.org/install/install_sources This is a good idea on AWS or bare metal, but won't make a difference on Rackspace due to them using an old hypervisor. If you're not compiling, install tensorflow from pip: pip install tensorflow If the device you're installing on has AVX2 or SSE4, install pillow-simd for faster image resizing: pip install pillow-simd if you only have SSE4, or CC="cc -mavx2" pip install pillow-simd if you have AVX2. I saw a significant increase in performance...