Hello image data: Train an AutoML image classification model

Use the Google Cloud console to train an AutoML image classification model. After your dataset is created and data is imported, use the Google Cloud console to review the training images and begin model training.

This tutorial has several pages:

1. Set up your project and environment.

2. Create an image classification dataset, and import images.

3. Train an AutoML image classification model.

4. Evaluate and analyze model performance.

5. Deploy a model to an endpoint, and send a prediction.

6. Clean up your project.

Each page assumes that you have already performed the instructions from the previous pages of the tutorial.

Review imported images

After the dataset import, you are taken to the Browse tab. You can also access this tab by selecting Datasets from the menu. Select the annotation set (set of single-label image annotations) associated with your new dataset.

Key point: An annotation set is the collection of annotations associated with a data type and a specific objective (image data type, classification objective in this case). For more information about annotation sets, see Creating an annotation set.

Go to the Datasets page

Begin AutoML model training

Choose one of the following options to begin training:

• Choose Train new model.

• Select Models from the menu, and select Create.

1. Go to the Models page

2. Select Create to open the Train new model window.

3. Select Select Training method, and select the target Dataset if they are not automatically selected. Make sure the AutoML radio button is selected, and then choose CONTINUE.

   

4. (Optional) Select Define your model, and enter the Model name. Click CONTINUE.

   

5. Select Train options. Select a model option according to your accuracy and latency needs. Optionally, enable incremental training and click CONTINUE.

   Incremental training considerations follow:

   • Incremental training can be enabled when there is at least one base model that has been trained in this project with the same objective.
   • Incremental training lets you use an existing base model as a starting point to train a new model rather than training a new model from scratch.
   • Incremental training generally helps training to occur faster and saves training time.
   • The base model can be trained from a different dataset.

   

6. Select Compute and pricing. Specify a node-hour budget of 8 node hours. Select Start training.

   Node-hour budget is the maximum time (may vary slightly) that the model spends training. This value is multiplied by the price per node hour to calculate to total training cost. More training hours results in a more accurate (up to a point) model but results in a higher cost. For development purposes, a low budget is fine but for production it's important to strike a balance between cost and accuracy.

Training takes several hours. An email notification is sent when the model training completes.

What's next

Follow the next page of this tutorial to check the performance of your trained AutoML model and explore ways of making it better.

Follow Deploy a model to an endpoint and make a prediction to deploy your trained AutoML model. An image is sent to the model for prediction.