Advanced Usage

In this section, we will cover some more advanced options in the image classifier toolkit that let you do more.

Annotating Data

If you only have images without corresponding labels, you can use the annotation utility built into the image_classifier. An example of its usage is shown below:

import turicreate as tc

# Use the example provided in the `Introductory Example`

# Use the Annotation GUI to annotate your data.
annotated_data = tc.image_classifier.annotate(data)

This utility will only present rows from the SFrame that have non-missing values. All missing values won't be shown in the GUI but they will be still included in the result set with default annotation value (undefined value, which denotes missing annotation). Exception will be thrown if no valid image is found.

If you forget to assign the output of your annotation to a variable, we've included a method to help you recover those annotations. The code for that is shown below:

import turicreate as tc

# Use the example provided in the `Introductory Example`

# If you forget to assign the output to a variable
tc.image_classifier.annotate(data)

# recover your annotation with this method
annotated_data = tc.image_classifier.recover_annotation()

The annotation utility supports label types of str and int.

Changing Models

The image classifier toolkit is based on a technique known as transfer learning. At a high level, model creation is accomplished by simply removing the output layer of the Deep Neural Network for 1000 categories, and taking the signals that would have been propagating to the output layer and feeding them as features to any classifier for our task.

The advanced options let you select from a set of pre-trained models which can result in a model having various size, performance, and accuracy characteristics.

Using the following option, you can change to use squeezenet which can trade off some accuracy for a smaller model with a lower memory and disk foot-print.

model = tc.image_classifier.create(
               train_data, target='label', model='squeezenet_v1.1')

Using GPUs

GPUs can make creating an image classifier model much faster. If you have macOS 10.13 or higher, Turi Create will automatically use the GPU. If your Linux machine has an NVIDIA GPU, you can setup Turi Create to use the GPU, see instructions.

The turicreate.config.set_num_gpus function allows you to control if GPUs are used:

# Use all GPUs (default)
turicreate.config.set_num_gpus(-1)

# Use only 1 GPU
turicreate.config.set_num_gpus(1)

# Use CPU
turicreate.config.set_num_gpus(0)