CODEWITHSUNDEEP
coremlturi-create
Martin Muldoon
Martin Muldoon

Reputation: 3436

What models do Core ML and Turi Create use?

I'm taking a course on Apple's machine learning technologies. I just came across this paragraph:

Turi Create and Create ML are task-specific, rather than model-specific. This means that you specify the type of problem you want to solve, rather than choosing the type of model you want to use. You select the task that matches the type of problem you want to solve, then Turi Create analyzes your data and chooses the right model for the job.

My question is when you select a task like binary image classification, and Create ML / Turi Create selects an appropriate model for you, what models does it have at its disposal. Aren't there MANY models in the market that do this? Did Apple simply pick the one they thought was best?

Thanks!

Upvotes: 2

Views: 779

Answers (1)

Matthijs Hollemans
Matthijs Hollemans

Reputation: 7902

For image-based tasks, Create ML uses a model named "Vision FeaturePrint.Scene". This is baked into iOS and macOS. It's probably some kind of ResNete model but who knows... it's a secret.

For image classification, Turi Create lets you choose between SqueezeNet 1.1 and ResNet-50.

For object detection, Turi Create builds something similar to Tiny YOLO v2 / Darknet.

For activity recognition, it's a model that consists of a conv layer, an LSTM, and a couple of fully-connected layers.

The Turi Create user guide actually explains some of this in quite a bit of detail.

I haven't really looked at the other models, but you can find out for yourself by looking at the Turi Create source code -- notably here -- or by training a model and then examining it with Netron.

Upvotes: 5

Related Questions