Cortado can be used as a trainer, with which the user can create his or her own themes. These themes can be used within VizX2. Before training the theme, a large representative collection of material needs to be made, containing both positive and negative examples of the envisioned theme. This material is then used to establish the neural network for this specific theme. Once the network is sufficiently trained, it can be added to VizX2.
VizX2 will then calculate a score for all unfamiliar/unknowns footage during indexation. This score is a number between 0 and 1 and reflects the degree in which the material complies with a certain theme.
Let’s say we want VizX2 to identify footage of swinging children. We first select a collection of images containing positive and negative examples. Next, we add them to the neural network using the trainer.
When these photos appear in a case, they will be scored by approximation.
Score: 0.9
Score: 0.1
During the selection of material to train for a specific theme, a few things are important:
- You have to put together a representative collection of material with relevant negative and positive examples.
- Determining the amount of material needed to train a reliable theme.
- Training takes time.
- It is difficult to train interdependent parameters.
ZiuZ offers support with the training of the first themes during the commissioning of Cortado.
