Object Class Recognition supports three different types of recognizer. All three use neural networks but have different characteristics.
Generation1
recognizer. This was the sole type of recognizer in Media Server versions 12.3 and earlier. This type of recognizer is the fastest to train but the other recognizers provide equal or better accuracy and are much faster at run-time.Generation2
recognizers offer the best accuracy, but take the longest to train. The training time is impractical on a CPU, so Micro Focus recommends that you train the recognizer on a machine with a supported GPU. You can use a trained Generation2
recognizer with or without a GPU but Micro Focus recommends using a GPU for best performance. If you run recognition on a GPU, this type of recognizer is much faster than the Generation1
recognizer, which allows you to set much shorter sample intervals.Generation3
recognizers are the fastest at run-time, but accuracy is slightly lower than the Generation2
recognizer.Before you train a recognizer you can also choose how many training iterations to run. Increasing the number of iterations improves the training and results in better accuracy, but each additional iteration that you add has a smaller effect.
To find the optimum number of iterations, Micro Focus recommends that you start with a small number of iterations. Double the number of iterations each time you train, until accuracy is acceptable. As a general rule, good accuracy can be obtained by multiplying the number of object classes in the recognizer by 2000.
|