Labeling training data is the first crucial step towards any deep learning application. The quality of this labeled data plays a major role when it comes to the application's performance, accuracy, and robustness.
With the MVTec Deep Learning Tool, we are creating a comprehensively smooth deep learning experience for HALCON users. By developing our own solution, we make sure that its output can seamlessly be integrated into HALCON. This application also allows us to incorporate our extensive experience and expertise that we acquired while developing HALCON's deep learning algorithms which – naturally – involved a lot of labeling, too.
Working with the MVTec Deep Learning Tool
The current version of the Deep Learning Tool (0.3) provides labeling functionalities for HALCON's deep-learning-based object detection and classification.
Labeling for object detection is done by drawing rectangles around each relevant object and by providing information about their corresponding classes. Depending on the project's requirements, users can choose between labeling their data with axis-aligned rectangles, or with oriented bounding boxes. Labeling for classification is done by simply importing images and assigning them the corresponding classes. If the images are stored in appropriately named folders, they can also be labeled automatically during the import.
To speed up the labeling process, multiple users can work on different parts of the image data set. E.g., a user can fully label a subset of images or label just a single class on all images within a data set. These partial data sets can then be merged into one project via different HALCON dictionaries.
To help you label data for HALCON's semantic segmentation, we provide Deep Learning Helpers for download.
The Deep Learning Tool is available in English, German, Japanese and Chinese (Simplified) language.
Early Adopter Versions
In addition to regular releases, we also provide Early Adopter releases of the Deep Learning Tool. With these versions, we give interested users the chance to try out latest functionality and encourage them to provide us with valuable feedback.
These Early Adopter versions run stable and are tested (but not as extensive as a regular release). They receive full product support from our dedicated support team. However, they might not include all translations and documentation, and downwards-compatibility to previous regular versions cannot be guaranteed.