Creating a new model
You can create new models (algorithms) in Criterion AI by choosing one of the many model types available in our model zoo. The model zoo includes models suitable for a wide range of purposes, including image classification, image segmentation, anomaly detection, predictive maintenance and many more.
Explore all of the models the model zoo offers by navigating to the Models section and clicking on the Create a new model button. This will display all of the available models. Some of the models are public and are generally available for everybody while other models might be exclusively available for you.
Choose the model type you are interested in by clicking on it. This opens up the details page for that model type where you can learn more about how models based on this model type will work as well as some requirements for the model.
In the example below, we have opened up the model type named Image Classification.
Depending on the given model type, some settings may or may not be available for you to configure when creating a model based on the model type. You can learn about these settings in the Instructions section.
Create a new model based on your desired model type by clicking on the Create model from this type button. This opens up a wizard with a series of steps that each require your input.
First, you must enter a suitable name of the model. It is best to choose a name that makes it easy for you to remember the purpose of and identify the model in the future.
Second, you must select the dataset(s) that contain the data that the model should be trained on. Depending on the model type, the model may accept one or more datasets. To learn what your model requires, please read the instructions for the model type carefully. It is your responsibility to provide a suitable datasets or set of suitable datasets that comply with the required structure by the model. If you do not supply the model with one or more appropriate datasets (according the model type instructions), the training of the model will fail.
After having chosen one or more suitable datasets, you may be required to supply some settings that are needed for the model to train. In this example, the Image Classification model type needs some information on the size and format of the input images as well as whether or not the images may be rotated and/or flipped for data augmentation purposes. Often, the default values will be alright but you may have to change them.
Finally, you will see a summary of the information you have provided. If you agree with the summary, click on the Create model button to create the model.
The model will be created right away and you will be redirected to its details page.
This concludes the creation of models.