How to start deep learning
We provides you full-featured toolbox containing the many steps required for the entire start deep learning process. With the integrated, user-friendly development environment we gave you , all steps from image acquisition, data processing, preprocessing to inference can be carried out. There are no interface issues to worry about.
Deep Learning Process
The general process of deep learning classification includes the following four steps.
Prepare: Acquiring, Labeling and Reviewing Data
Acquire image data for deep learning under conditions similar or identical to those expected in real-world applications
Label every object in the dataset and every object in a class in the same way to ensure the correctness and accuracy of the labeled data.
Review and check for mislabeled data. Using free MVTec deep learning tools, you can prepare datasets easily and efficiently.
Training: Train Your Own Deep Learning Neural Network CNN
After exporting data from deep learning tools to HDevelop, HALCON can analyze these images and automatically learn which features can be used to label a given class. This is a great advantage over traditional taxonomy methods, where these features have to be "manually defined" by the user.
You can train your own classifiers based on the pre trained CNNs (Convolutional Neural Networks) included in HALCON. Based on hundreds of thousands of images, these networks have been highly optimized for industrial applications. In HALCON, it is also possible to use neural networks in ONNX format previously generated in third-party platforms.
Evaluate: Validate the trained model against the test data
To verify that the performance of a trained deep learning model is sufficient for your application, you can choose between various visualization options.
For example, you can use the confusion matrix in HALCON to get an accurate reading of the proportion of correct and incorrect samples. Heatmaps can show which regions in an image are particularly important for a neural network to make a decision.
Inference: apply the evaluated network to new images
Once the neural network has learned to distinguish between a given class, for example, identifying whether an image shows a scratch, contamination, or a good sample, the user can apply the newly trained CNN classifier to new images, which is called "inference." This inference can be performed on both the GPU and the CPU (x86 and Arm® based).
The following services will help you start your deep learning journey
Technical Training
MVTec and MVTec partners provide training to help you use deep learning in MVTec products. Feel free to check out our training and seminar program.
teaching video
Watch our instructional videos, and use our HDevelop sample programs, where you'll get plenty of explanation and background information.
0 Comments