The underlying base model used to extract image features, during the training of an image classifier model.
- macOS 10.14+
- Xcode 10.0+
- Create ML
When you train an image classifier with Create ML, you don’t train a complete model that converts images to labels. Instead, Create ML leverages a feature extractor, which is another model that identifies a set of general but distinguishing image characteristics. You can either use Create ML’s
scene feature extractor or provide your own custom feature extractor.
In either case, you train your model to map the feature extractor’s output to labels in a process known as transfer learning. Your model effectively borrows the knowledge of the feature extractor to accelerate its own training process while requiring fewer training images.