ML Image Classifier not giving expected results on simulators


I created an Image Classifier app, “Chichibu Unkai Watcher”, and it returns right results on a real iPhone device. Also it returns right results on mac catalyst. But, I found that the iPhone/iPad simulators do not return expected results.

XCode Version 13.1 (13A1030d)
Simulator Version 13.1 (970)
Create ML Version 3.0 (78.6)

Fig.1 iPhone real device: Unkai confidence level is ~0%.
Fig.2 mac app on macOS Catalyst: Unkai confidence level is ~0%.
Fig.3 iPhone Simulator: Unkai confidence level is 69%
All of those are using a same image data, and the Unkai confidence level should be ~0%.


I tried some GPU settings on the simulator app, also some compiler settings changes, but it does not resolve this symptom. I found an article on the Stack Overflow which says it might be caused by the simulator. At this moment, I’m assuming this is caused by the simulator.


Per Sheldon, he is suggesting to run one on a real device.

Fig.1 iPhone real device

Fig.2 mac Catalyst app run on a real mac

Fig.3 iPhone Simulator