Model Training

This section provides comprehensive information on various methods to generate the model, how to initiate and monitor the training job, and the steps to download the model files from the Imagimob Cloud.

Generating model

After you collect the data and design the preprocessor, the data is passed to the model for training. Before the model can be trained, you need to generate the model and define the layers of the model. Studio allows generation of multiple different models to be trained and compared to find the best fitting model.

You can generate the model using any of the following ways in Studio:

• Model Wizard: You can generate multiple model architecture using the Model Wizard. The Auto ML tab in wizard can be configured for different model families, classifiers, sizes, and learning rate. The wizard generates models based on the configuration, speeding up the model development phase by prioritizing the main features of a model. The generated models have different layers and layer configurations and produce slightly different results. The models generated by the Auto ML wizard can be changed layer by layer, allowing the more advanced users to fine tune the models. Refer to Generating model to know about Model wizard.

• Empty model: You can create a custom model from scratch or clone one of the Auto ML models as a starting point for a new model. For each model in the list, you can change the training settings such as the number of epochs to be run and the learning rate used by entering the desired values. You can click the Model properties icon to set additional properties. If you select a model, you can view the different layers of that model. You can change the order of the layers, properties of each layer, and add or remove layers before training.

• Import architecture: You can import an existing Keras H5 model in different modes into the Studio to generate a model.

Refer to Generating model for more information.

Starting and Tracking Training Jobs

Once the model is generated, you can initiate and monitor the training job. At this stage, you have the option to use either GPUs or the default CPUs for training. After starting the job, all data and model information will be uploaded to the Imagimob Cloud for the actual training. Refer to Starting and Tracking Training Jobs for detailed steps.

Downloading Model Files and Import predictions into your Project

After the model is trained in the Imagimob Cloud, the next step is to download the files and import the model predictions as tracks into the sessions containing the original data. This provide a much more detailed view for evaluating the performance of the model. Refer to Downloading Model Files and Import predictions into your Project for detailed steps.