AI services
The use of AI has become a must in computer science. Its detailed understanding requires a considerable mathematical background. Many parameters need to be carefully set to achieve optimum performance. Keur Studio has AI experience in training image classifiers and object-in-image detectors.
In the case of a classifier, the aim is, given an image and a list of categories, to automatically determine which category the image belongs to. To do this, a mathematical model is trained to identify the recurring patterns specific to each category, based on a large number of examples (from several thousand to several million) of each of the categories we wish to distinguish, so as to be able to recognize them on new images and thus identify the right category.
In the case of a detector, the aim is to determine all the objects present in an image, their categories and their positions (generally with a rectangle enclosing the object). The associated model is generally more complex to train than in the case of classification, as it must be able to distinguish several objects within a single image. Once again, the creation of an annotated annotated database is a crucial step.
Contact usA tedious but essential step, annotated data must be available in large quantities to guarantee the best possible performance.
Starting with a list of the categories you wish to distinguish or detect, Keur Studio can build up an appropriate dataset by uploading images for each of your categories and annotating them. This image database is sized according to your needs.
In the case of a classification objective, annotation simply consists of indicating to which category each image belongs. In the case of a detection objective, the annotation includes, for each image, all the objects present, the category of each of these objects and a rectangle enclosing each object.
Numerous models exist, each with its own specificities, whether in terms of the volumes of training images required, inference time or performance. In addition, the models used for classification differ from those used for detection.
Keur Studio determines the models best suited to your problem and trains them on the dataset to determine the most efficient. As an output, you'll be provided with minimal functional code, as well as a file containing the trained mathematical model, enabling you to perform inference on an example image. Now all you have to do is integrate this trained model into your project!