Training Fusion Models

This page will provide a walk-through for training Fusion models in EdgeFirst Studio.

Note

Checkout our full video tutorial above as part of the EdgeFirst Studio Series to showcase the steps for running Fusion training in EdgeFirst Studio. Otherwise, follow along the steps shown below with section specific timestamps of the video.

Verify Dataset

Before running a training session, ensure the dataset is ready to be used for training. This means that the dataset is properly annotated and the dataset is properly split with training and validation samples.

The sample dataset shown below has a dedicated split for training (20066 samples) and validation (2229 samples).

Dataset Groups

To verify the annotations, click the button that navigates to the gallery. This will show the contents of the dataset. The dataset may be comprised of multiple sequences as shown below.

Dataset Sequences

Clicking on any of these sequences will open individual images in the sequence with the visualizations of the annotations.

Info

Datasets that train Fusion models provide world annotations of the object's 3D bounding box. For more information on the dataset annotations, please see EdgeFirst Dataset Format.

Fusion Annotations

For cases where the annotations need corrections, please see Dataset Tutorials for more details.

Select the Trainer Tool

Select Trainer from the tool options.

Tool Options

Specify the Project

Specify the project to run training at the center of the top menu bar.

Project Selection

Create Training Experiment

If you haven't already done so, create a training experiment. Create a new training experiment by clicking the create button on the top right.

Create New Experiment

This will provide pop-up for the user to specify the name and description of the experiment. Give a name and a description that reflects your goals in this experiment.

Create New Experiment

Create Training Session

Create a new training session within this experiment by clicking the NEW SESSION button as shown below.

Training Session

Configure the settings on the left panel by specifying Trainer Type to EdgeFirst Fusion and provide additional configurations for the name of the session and the dataset to deploy. Next configure the settings on the right panel by specifying training parameters. By default the Fusion model is configured with both camera and radar inputs, however, a Camera-Only or Radar-Only model are possible variations.

Note

Additional information on these parameters are provided by hovering over the info button. For more information on available vision augmentations please see Vision Augmentations.

Training Options

Start the Session

Start the session by clicking the START SESSION button on the bottom right.

Start the Session

Session Progress

The training session has now started while the progress is tracked on the left panel and additional information and status is shown on the right panel.

Training Session

Training Session Fields

Training Metrics

The training metrics are shown by clicking the button that views the training charts on the top left of the session card.

Training Metrics

Completed Session

Once completed, the status will be shown as complete.

Completed Session

Trained Models

The trained Keras and TFLite models can be found and downloaded by clicking on the button the views the session details on the top right of the session card.
This will open a new dialog with the session details and the models are placed on the top right which can then be downloaded.

Session Details

Next Steps

Now you have generated your Fusion model, follow these next steps for validating your Fusion model.