Developing industrial deep learning models

Features & Workflows

The MVTec Deep Learning Tool combines all the essential steps for developing industrial deep learning models in one software. The user interface is clearly structured and allows for working without programming.

ONE TOOL FOR ALL STEPS

Developing Deep Learning Models

1. Labeling

2. Training

3. Evaluation

4. Data Management

5. Deployment

DEEP LEARNING TASKS

1. Labeling (Annotation)

The MVTec Deep Learning Tool offers several annotation methods for different deep learning tasks.

Classification

Labeling for classification is done by simply importing the images and assigning them to a class. If the images are stored in appropriately named folders, they can also be automatically labeled during import.

Please note: Once you watch the video, data will be transmitted to Youtube/Google. For more information, see Google Privacy.

Object Detection

In object detection, labeling is done by drawing rectangles (bounding boxes) around each relevant object and assigning these rectangles to the corresponding classes. Depending on the project requirements, the user can label the data either with axis-aligned or oriented rectangles.

Instance & Semantic Segmentation

Draw polygonal regions around relevant objects or create pixel-precise masks with a brush and eraser that cover the relevant objects.

In addition, several smart labeling tools are available to accelerate annotation. These tools generate instant annotation suggestions – either after selecting a relevant image area or when moving the mouse pointer over an image area.

Please note: Once you watch the video, data will be transmitted to Youtube/Google. For more information, see Google Privacy.

Deep OCR Training

Re-training a Deep OCR model can significantly improve the recognition rate of Deep OCR in MVTec HALCON for specific applications. The Deep Learning Tool supports the efficient annotation of large datasets through optimized workflows. Detection and recognition parameters can be configured to generate automatic annotation suggestions, which can be adopted or refined, significantly reducing manual effort.

Anomaly Detection

Global Context Anomaly Detection

This method highlights atypical image content based on the global image context and requires minimal annotation effort. Images are classified as "Good" or "Anomaly," enabling quick and efficient labeling. This simplifies the identification of anomalies on a large scale without the need to annotate every image in detail.

DEEP LEARNING TASKS

2. Training

Training for classification in the Deep Learning Tool.

During training, a pre-trained classifier is trained on the previously labeled image dataset. With each iteration over the training dataset, the model attempts to improve its predictions on the validation dataset. Based on the results, the weights within the neural network are adjusted to improve performance in the next iteration.

In the Deep Learning Tool, users can configure all relevant training parameters. After selecting a data split, training can be started, and progress and performance are visualized.

Training is currently supported for the following deep learning methods:

Classification
Global Context Anomaly Detection
Object Detection
Instance Segmentation
Semantic Segmentation
Deep OCR (Detection and Recognition)

DEEP LEARNING TASKS

3. Evaluation

User interface of the MVTec Deep Learning Tool during evaluation for classification.

During evaluation, the model is assessed using the test dataset. This step provides an indication for the image processing specialist of how well the model is expected to perform in practice.

Users can evaluate and compare their trained networks directly within the tool. The Evaluation section provides information about model accuracy, including a heatmap for the predicted classes of all processed images, as well as an interactive confusion matrix that helps identify misclassifications. Users can also calculate the estimated inference time per image and export the evaluation results as a standalone HTML page for documentation purposes.

Evaluation is currently supported for the following deep learning methods:

Classification
Global Context Anomaly Detection
Object Detection
Instance Segmentation
Semantic Segmentation

DEEP LEARNING TASKS

4. Data Management

The MVTec Deep Learning Tool supports the structured organization of image data and projects throughout the entire deep learning workflow.

Management of training and image data within projects
Organization of data for annotation, training, and evaluation
Loading and saving of projects
Consistent use of datasets within a project for training and evaluation

5. Deployment

Trained models can be exported and used within the MVTec product ecosystem:

Export for MVTec HALCON for programmatic integration.
Export for MVTec MERLIC for use in no-code machine vision applications.

DOWNLOAD

Get started with the Deep Learning Tool

The MVTec Deep Learning Tool offers a fast path to a complete deep learning solution with its intuitive user interface. It provides active support for optimizing trained networks, ensuring the best performance. The tool integrates seamlessly into the MVTec portfolio, allowing for smooth integration into existing workflows. Additionally, it gives you full control over your own data, ensuring privacy and flexibility throughout the process.

Go to downloads