How to Manage Images in In-Sight ViDi

Image management is critical to deep learning applications because these applications may use hundreds of images taken from a production line. Images must capture the most common production scenarios that a machine vision camera will encounter as parts or components move through the process. The image set must reflect as many of these scenarios as possible.

Cognex In-Sight ViDi tools have plenty of options to help you manage the images that are the bedrock of your machine vision application. This video walks you through the image-management features and illustrates the importance of:

Folders. The images that train your deep learning application have to be divided between 'good' and 'bad.' Good images represent the proper state for a part or component. Bad images represent defects like scratches, dents, or fluid spills. Separating the images into folders named 'good' and 'bad' simplifies this process.

Labels. Adding labels to a filename makes it easier to find all the images that share a common characteristic. Labeling can separate images into groups in situations where you might not want to place them in folders. A label might be 'good' or 'bad,' but it also might refer to any number of variables - like the time of day the image was taken.

Sorting. In-Sight ViDi helps you group similar images and complete tasks like applying a label or adding to a folder. When you're dealing with huge image sets, sorting accelerates everything. Sorting also ensures that your application gives equal weight to all of your training images. Without sorting, the algorithm might give too much weight to one fraction of the images and not enough to another. Sorting can balance out the distribution.