Lets paint a scenario:
4 folders containing 100 images each. One went to IBM Watson for training using NodeJS, the other set went through TensorFlow using the process from the Scavenger Hunt demo (
https://github.com/google/emoji-scavenger-hunt).
IBM Watson is all API based so it was pretty straight forward. Upload to their system and wait.
TensorFlow on the other hand, I used python to train the same set of images and just used my own computer (MacBook 2015 edition maybe).
Can we assume that IBM Watson might be backed up with a good server farm and algorithm better than tensorflow thus creating a faster and more accurate result? From my initial testing of a Dandelion turned out to be a Thistle