We are proud to introduce an all new desktop version of the alwaysAI Model Training Toolkit. The new robust interface and simple workflow of the desktop app will allow alwaysAI users of all levels to train their own Computer Vision models. The model training desktop application is the first of a suite of applications that will support your Computer Vision model operations, from data-collection and organization to model deployment and maintenance. The model training module is the backbone of this toolkit and now it has a new look. Let me walk you through it.
First things first: download the latest installer for your system here. If you already have alwaysAI installed, simply look for the bar at the top of the application that indicates there is an update available (Figure 1).
*Note: if your version of alwaysAI was installed before 11/13/2020 you won’t see the upgrade bar, so download the latest version from above link.
Once the update completes, click the now-available “Model Training” button in the Train Your Own Model section. This will open the model training interface (Figure 2). You’re ready to go.
The new interface has three main sections: the configure pane on the right, the main model training area front and center, and below that is where the training summary is displayed once your model has finished training. The configure pane is where you will do most of your work. Upload a dataset (.zip file of images and annotations) from your computer. If you don’t have one we have a couple sample datasets here and here. Once your dataset is loaded you can see all the labels identifying the objects your model will be trained to detect in the Labels section. Give your model a name and then configure your model training session by setting the number of epochs to train and the batch size to use. *Note: These downloads are meant to be .zip files. If your browser automatically unzips upon download (Safari) you will need to compress the “Annotations” and “JPEGImages” folders together into a .zip.
Once you have configured your training session, press the “Start Training” button at the top of the page. We’ve optimized the hyperparameters for general object detection so you can get a decent model without the need for advanced configurations. We plot the loss for each step, and the validation loss after every epoch. Look in the “Logs” tab for more details on the performance of your model.
When your model finishes training a training summary section will appear under the chart. Here you can find details on how the training session went, as well as instructions on how to use the model. Deploy the newly trained model with a few simple commands. Use your model as you would any model from the alwaysAI Catalog.
As part of the model training team at alwaysAI I am excited for the potential of this new desktop interface for two reasons. First, the modern technology framework will allow us to rapidly iterate and add features to our model training tool. Second, we can build interfaces for the other steps of model operations, starting with creating and managing datasets, and eventually incorporating all the computer vision model deployment and management capabilities that alwaysAI is known for. Stay tuned, the future holds great things!