CreateML what kind of ObjectDetector Network is trained?

Question

I used CreateML do train a new custom ObjectDector. Everything worked well so far.

Now I am just wondering, what kind of Network is trained in the background? Is it something like YOLO or Mobilenet?

I did not found anything on the official documentation:

score 2 · Accepted Answer · answered Sep 07 '20 at 09:39

2

There are two options:

TinyYOLOv2
Using transfer learning. This uses a built-in feature extractor model (VisionFeaturePrint.Objects). This is available with Create ML in Xcode 12.

answered Sep 07 '20 at 09:39

thanks for your answer. where can I choose between those options? – simplesystems Sep 07 '20 at 16:04
When you train an object detector in Create ML you can use to enable transfer learning. This option is only available in the version of Create ML that comes with Xcode 12 (currently in beta). The trained model will only work on iOS 14+. – Matthijs Hollemans Sep 07 '20 at 19:53
ok cool! and why do you know that the default is tiniyYOLOv2? do you have a source? – simplesystems Sep 08 '20 at 07:23
I looked at the generated model in Netron. – Matthijs Hollemans Sep 08 '20 at 09:24
ah ok, how do you identify which modell it is in netron? – simplesystems Sep 08 '20 at 10:39
I know what TinyYOLO looks like, i.e. what layers it uses. I think they may also have mentioned that it uses YOLO in this year's WWDC videos about Create ML. – Matthijs Hollemans Sep 08 '20 at 11:49

1 Answers1