Highlights of the AI-Xtension 6.2.6 (6.1.8)
PI_NeuralNetworkTools 6.2.6 (6.1.8)
1) Anomaly Detection
The new check function Detect Anomalies provides a way of using Anomaly Detection networks in NeuroCheck. The key advantage of Anomaly Detection is that training these models only requires few Not-OK samples in order to distinguish between good and bad parts. This Deep Learning discipline thrives especially when production processes are stable and only few Not-OK samples exist.
2) Performance Improvements and Hardware Support
By applying further parallelization and optimization, we could improve overall performance of the plug-in up to 50 % compared to the previous version. This performance gain is especially visible in CPU applications. Hardware used for inference is now configured directly via the check function parameter dialog. The plug-in supports NVIDIA GPUs with Compute Capability from min. 3.5 to 8.0.
3) Legacy Support
It is a common use case to upgrade existing machines with Deep Learning solutions to boost performance. Consequently, PI_NeuralNetworkTools is available for NeuroCheck 6.1 and Windows 7 as well. This way it is still possible implementing Deep Learning solutions into existing machines without any hardware changes.
4) TensorFlow Support Cancelled
Recent progress in Deep Learning still happens at a high pace. The former market leader google TensorFlow has lost a large part of its community during these developments. Some bad decisions in design led to major performance flaws and drastically increased efforts for maintenance and care. ONNX in the meantime, offers a powerful alternative with an ever-growing community and excellent performance. We hence decided to cancel TensorFlow support in our plug-in. Please see Migration of Legacy Projects on how to transfer existing TensorFlow models from our Toolkits to ONNX.
5) ‘Advanced‘ Check Functions Removed
Customer feedback and our own experiences from projects showed, that the use cases for the ’Advanced’ variants of the plug-in check functions are very limited or solvable otherwise. Based on these insights we decided to remove these check functions in order to reduce complexity and effort for maintenance and care.
NcDLToolkit 1.0 – One Toolkit for Everything
NeuroCheck Toolkits for creating, training and exporting Neural Networks were revised completely. The Deep Learning tasks Classification, Object Detection, Anomaly Detection and Segmentation are now available in one single Toolkit. Configuring the training, evaluation and export pipeline was simplified as well.
Deep Learning Trainings
NeuroCheck Deep Learning Toolkit is still and will remain free of charge. However, we recommend our Deep Learning Training Course in order to successfully implement AI projects in general and using the Toolkit in particular.
For dates and individual contents, please contact our support-team.
Migration of Legacy Projects
Importing projects from previous versions of the Toolkit is possible via copying the dataset into the new Toolkit and adjusting training parameters accordingly. After a new training run you can export the resulting model to ONNX and use it with NeuroCheck.
In case of any questions or problems regarding the migration process, please contact our support-team.
Note: For using 3D Xtension requires an extension of the NeuroCheck license.
For questions or feedback, please contact our support team.
Copyright notice for the photos used on this page:
Images © NeuroCheck