Training the Model to Improve Accuracy


#1

I’m really liking retrobatch. Great work.

For me one of the most intriguing feature is the classification node.

Is it possible to further train the models that are used? How does one train the model in Retrobatch?

Reason: When I run the model some photos are mis-classified. I’d like to train the model for higher accuracy.


#2

Training an existing model isn’t something that’s possible with Retrobatch. However, you can make your own models with some open source tools like Turi Create: https://github.com/apple/turicreate

When MacOS Mojave 10.14 is released this fall, there will be new things in there for Retrobatch to take advantage of for making your own classifications. I don’t know exactly when I’ll add those to Retrobatch, but I hope it’s sooner rather than later.


#3

Thanks for the info, Gus. Sounds like there are some exciting developments coming up.

Really loving the app.


#4

I’m also interested in training a model of my own. Please share your experiences here @highq


#5

@gingerbeardman sure will.