Open yrik opened 6 years ago
This is a great feature request!
Right now the only way to do this is by training a binary classifier (using the image classifier toolkit) for each label separately. Of course, training time does not scale well with the number of classes and even worse the Core ML deployment would not give you a single model that would allow you to do inference on all classes at the same time. To consolidate the models would require some serious coremltools hacking. It would be great if Turi Create could make this easy.
Even, i need to classify multiple images. i implemented this by using following code
data["chairType"] = data["path"].apply(lambda path: "Base" if "base" in path else ("Chair" if "chair" in path else "Roller"))
Its not classified property and i tired with four/five classification of the images but i didn't workout. i have all the image set on the respective folder
+1. This blocks many potential use cases for once the produced models are deployed on-device. For computation on non-iOS devices, it is slightly less of a blocker, but a large pain nonetheless.
Is there any plans to support multilabel classification? I miss this feature badly..