SynthAI Demo #3: Synthetic data model testing and evaluation [VIDEO]

By Kelly Gallagher

As the use of artificial intelligence in industrial settings continues to grow, the need for effective methods of training and deploying machine learning models becomes increasingly important. Traditional approaches to data collection and labeling can be tedious, costly, and time-consuming. These challenges make it inherently difficult for engineers to keep up with the demands of rapidly evolving industries, such as the recent use case with a leading electronics plant where SynthAI helped to accelerate the set-up time for AI inspection models five times faster than before.

By leveraging a self-service synthetic data generation platform like SynthAI, engineers can quickly and cost-effectively generate high-quality data to train and deploy reliable machine learning models. In our previous product demonstration, we showed you how SynthAI combines synthetic data with real images to create a hybrid approach that allows for enhanced model accuracy in production environments.

Now, we invite you to join us for a demonstration on testing and deploying trained models using SynthAI. You’ll learn how to iteratively improve the accuracy of your machine learning models with high-fidelity synthetic data, streamlining your modeling pipelines and helping you to stay ahead of the curve in the competitive world of industrial AI. Follow along to learn how to test and deploy a trained model so you can continue to enhance the accuracy of your ML models with quality synthetic data in the least amount of time.

Download the trained model to a local folder:

After the fine-tuning training session has been completed, you can download the trained model to a local folder and begin to test it by selecting “Download Trained Model.” You can also download the Synthetic dataset by selecting the “Download Synthetic Data” button. 

Familiarize yourself with the “Trained Model Package”

Open the Trained Model folder, which contains files, including:

  • Objection detection model
  • Model configuration
  • Statistics

In this folder, you can find guides and files that will help you test the model, for example a README file, Inference Snippet python script, and an example image. For an initial test of the trained model, open the “README” file and follow the instructions. Finally, you can activate the virtual environment and run the code.

View the Model’s results

To evaluate the model’s results, open the image including the predictions and scores. Here, you can view the annotations that have been created over the image and the level of confidence associated with each prediction.  

Replace the “Sample Image” with a real image to test:

Alternatively, you can replace the “Sample Image” with another image that represents a “real object” in a physical environment and run the test again.  

Model improvement:

If you want to further improve the model’s precision, conduct another fine-tuning session while adding more representing images of the object you are training the model on. Interested in learning how to use the fine-tuning tool? Watch the previous training that demonstrates how to use this capability.

By leveraging the power of SynthAI, engineers can overcome the limitations and expenses of acquiring real-world data and instead utilize high-quality synthetic data to train and optimize their machine learning models. The integration of AI in industrial use cases has the potential to revolutionize manufacturing operations by increasing efficiency, reducing costs, and improving product quality.

Interested in learning more? Visit our Tecnomatix Community for an in-depth, step-by-step training of the product demonstration. Unlock the secret to revolutionizing your industrial operations with SynthAI – the cutting-edge technology that’s taking AI-powered computer vision systems to the next level.

Leave a Reply

This article first appeared on the Siemens Digital Industries Software blog at