 Deep Learning Tool 25.04 from MVTec provides developers with a fast path to the complete deep learning solution. Its intuitive user interface ensures high-quality labelled data which is crucial for accurate and robust deep learning applications. The tool integrates seamlessly with HALCON and MERLIC, supporting a wide range of tasks including object detection, segmentation, anomaly detection, and Deep OCR.
[Request a Professional Demo]
MVTec’s Deep Learning Tool 25.04 is offered in three tailored editions to meet a wide range of development and deployment requirements:
Deep Learning Tool (Free Edition):
Ideal for initial experimentation and smaller projects, this edition provides core labelling, training, and evaluation functionalities with a user-friendly interface.
[Try the Free Edition today]
Deep Learning Tool Professional Edition
Designed for professional developers and integrators, this edition adds advanced features including enhanced dataset management, extended model configuration options, and long-term compatibility with HALCON and MERLIC. It supports larger-scale projects.
[Learn more about the Deep Learning tool Professional Edition]
Deep Learning Tool OEM Edition
Tailored for original equipment manufacturers embedding AI into their products, this edition offers extensive customisation and integration capabilities. It enables seamless embedding of deep learning workflows into OEM solutions, with licensing options suited for high-volume deployment
[Contact Us About OEM Licensing]
Labelling, Training, Evaluation
MVTec’s Deep Learning Tool is organied into three core areas, Labelling, Training, and Evaluation each serving a critical role in your workflow: you begin by accurately annotating your images in the Labelling workspace, move on to fine-tune network architectures and parameters in the Training environment, and finally validate model efficacy and identify improvement opportunities with the comprehensive Evaluation dashboard. |
Labelling
Data labelling is an essential task in many deep learning projects, providing the system with essential information about how to solve the problem correctly. This information may include image classes, object locations, or pixel masks assigned to classes or instances.
Labelling for Classification
Import images and assign them to classes. If images are organised in correctly named folders, they can be labelled automatically during import.
Labelling for Object Detection
Draw rectangles around each relevant object and assign these to corresponding classes. Label with either axis-parallel or oriented rectangles depending on project needs.
Labelling for Segmentation
Create polygonal regions or paint pixel masks using brush and eraser tools to cover objects for semantic and instance segmentation. Smart labelling tools offer instant suggestions, speeding up the process.
Labelling for Deep OCR Training
Retrain Deep OCR models for improved recognition in specialised applications. The tool offers workflows that automatically generate label proposals, which can be accepted or refined easily.
Labelling for Global Context Anomaly Detection
Simply import images and assign them to "good" or "anomaly" classes. Automatic labelling during import is available when images are properly organised.
Training: Controlled, Transparent and Flexible
During training, a pretrained classifier learns from the labelled dataset by iterating over the data to improve predictions based on validation results. The neural network weights are adjusted to enhance performance progressively.
Users configure key parameters on the training page, select data splits, and monitor training progress with real-time visualisation of performance metrics.
Training currently supports the following deep learning methods:
• Classification
• Global Context Anomaly Detection
• Object Detection
• Instance Segmentation
• Semantic Segmentation
• Deep OCR (Detection and Recognition)
Evaluation: Comprehensive and Insightful
Evaluation tests the trained model against a test dataset to predict real-world performance. It provides machine vision specialists with critical insights into model accuracy and reliability.
• Users can evaluate and compare trained networks directly within the tool.
• Features include heatmaps for predicted classes, an interactive confusion matrix to identify misclassifications, and estimation of inference time per image.
• Evaluation results can be exported as an HTML report for documentation.
Supported evaluation methods include:
• Classification
• Global Context Anomaly Detection
• Object Detection
• Instance Segmentation
• Semantic Segmentation
Seamless Integration
The Deep Learning Tool seamlessly integrates into the MVTec product portfolio with HALCON and MERLIC and serves as the core of your Deep Learning application.
Acquire your images and preprocess them with HALCON or MERLIC if necessary. After labelling, training and evaluation in the Deep Learning Tool, deploy your trained network in the respective runtime environment.
[Request a Professional Demo]
|
info@adeptturnkey.com.au
Follow us on LinkedIn
