Pixel-wise labeling (classification)
Pixel-wise labeling allows defective areas on the images in Classification projects to be labeled to improve inspection accuracy. It is possible to achieve high inspection performance even with insufficient image data.
Network options (segmentation)
A “contextual network” is a type of segmentation that minimizes overkill rates compared to “sensitive network” that minimizes the underkill rates. By choosing one of these two networks, you can optimize your defect detection results.
Visual debugger (classification)
This visual debugger feature visualizes which areas of the image the deep learning algorithms has focused in on and analyzed. It allows the user to check whether the inspection has been carried out according to the user’s intention.
Continual learning (single x classification)
Continual learning is a way to train the deep learning model using the pre-trained neural network in order to minimize the training time and the number of training images required for each new product.
Uncertainty data analysis
This function provides an easy way to review training data. It allows the model to analyze and extract vague data that is difficult to classify as OK/NG and easily finds falsely classified images caused by incorrect labeling.