Create ML templates - only see image and sound

Im using Xcode 11 and CreateML 1.


Only see image and sound...
Where are other templates?




Replies

I am looking for the other templates as well. I suspect (hope) Create ML will be updated in a week or 2. I'd like the Object Detector image of the Image Classifier.

Yup, hope Apple launches with new CreateML update.
Do you know how to work with existing .mlmodel files in CreateML?

I suspect Apple will come out with a new Create ML upon the next update. I emailed a number of the ML engineers to try to find out what is going on.


I've not used external .mlmodel files before but there were instructions I saw somewhere on the developer site. A concern that one of the Apple Engineers expressed in one of the videos was to beware that these public models are pretty generalized so don't work ideally in actual implementations - or better said: they are great starting points but should be amended to or replaced by your own data collection efforts.


One interesting thing to note is that while Apple provides a great tool in Create ML, they "forgot" to design at least a basic image collection gathering toolset (that I am aware of) to help one gather and tag images in an easy manor like Create ML. Things can be done in Python and Playgrounds but it would have been nice for them to provide a comprehensive toolset for image (and audio) gathering and tagging.


Still, Create ML and Apple's ML to begin with is AMAZING!

BTW, here are Apple's community derrived models: https://developer.apple.com/machine-learning/models. A decent starting point perhaps.

Yup, I’ve looked at those models also. Do you know how to custom train one of those models with your own datasets?

I think you mean, how do you add to one of these existing datasets using it as a base, right? No I don't - that's a good question. I do remember watching a video on how a user can modify an existing dataset and customize it with their own data, but I don't know how to take a pre-baked dataset and add to it during development then deploy that amended data in a final app. But there must be a way!


Sorry...

Looking for Activity/Motion templates in Create ML similar to what was shown in the sessions with the Frisbee. If anybody knows when we will see these templates, please let me know. Would like to start playing with these asap. 😀

Take a look at https://developer.apple.com/videos/play/wwdc2019/704/ to embed your users own data back into the model.

You can look for public models to start. https://developer.apple.com/videos/play/wwdc2019/704/ talks a little about that at the beginning.

Well well, xcode beta 2 released and no new templates..

Yes, no new templates. Hmmm.... I wonder when.


FWIW, I wrote a playground that calls the MLObjectDetector framework and while it works and results in models, it is extreamly slow to produce even modest models - not sure how much it is actually using the GPU, and the results out of these models so far are super sub-par in terms of object detection. I am still working on that. It makes me think there might be bugs deep down inside the Object Detection guts, but I don't know for sure. Maybe when they fix these supposed bugs (mid June of 19 now), the templates will show up...

Yeah, i think it will be released when it holds what we got promised in the session.. im still pretty excited!

Yeah I think you’re right. I’ve been an Apple developer since the Mac first came out in 1984 and honestly this is one of the most exciting updates / releases I’ve seen in 35 years with Machine Learning, Augmented Reality, new programming methodology in Swift, and a ton of other things. Very exiting future!

Still nothing, nearly three weeks later.

Yeah, it's kinda how Apple does things. In a way it makes sense - work out as many bugs as possible then do a big release. But in the mean time, us developers are kinda in a holding pattern. I'm sure they know that and have a release schedule, so we'll just have to wait. I think they said that iOS will go public beta in July so we should definitly see something before that release.