The Mav image classification service uses Computer Vision for OCR and Tensorflow. The primary purpose of Mav is to find alt tags for images using base64 data.
Some possible env configurations you may want to use:
# saves tons of mem when set to true at least 500mb of extra memory required to enable.
# Computer Vision
# PageSpeed or lighthouse
This service uses a couple different types of networks to determine if an image alt is valid.