Software specifically for object annotation for macOS’s CreateML
Recently, I’ve been working on implementing object detection functionality on iOS, specifically to train a model for fuel dispenser information extraction. Since it’s an iOS project, my first choice was to use Create ML, which is built into macOS, to train the model. However, after looking at several mainstream data annotation tools, I found that most don’t directly support Create ML. After refining my search keywords multiple times, I discovered the native macOS app LabelML, which directly supports Create ML training datasets.
The functionality is not complicated, and the app has native interaction with fast speed. If it's just for simple object annotation, I would highly recommend it—it’s practically ready to use out of the box.
- Once you open the app, you can directly select a folder—no extra pages.
- Interaction is convenient (similar to Photoshop interaction):
- Scroll the mouse = Zoom in and out of the image.
- Hold the spacebar = Drag the image with the mouse.
- Command + S to save anytime (native interaction).
- Command + Z to undo actions (native interaction).
Since the functionality is relatively simple, here are some screenshots of the usage:
-
Click the menu to activate folder selection, ensuring that the folder contains images:
-
Annotation:
-
Create a new ObjectDetector in CreateML:
-
Choose the folder that you just annotated in the TrainingData:
-
Click the 'Train' button in the top menu to start training:
-
Once training is complete, drag the trained model into your iOS project for integration.
