Sources

Use the Collection>Sources view to see the data sources that are available for you.

Required Steps

No required steps. You can use the default data sources that are available:

• a default source of the mode "Train Data for Recognition"

• a default source of the mode "Train Data for Background"

Note:

If your project has the use case Classic Classifier, and you use the default source for “Train Data for Recognition” or "Train Data for Background", a subset of the images (~10%) will be automatically split out to a "Linked Test Data" source in order to make sure that you have test data available. You can see if a source has linked test data under the field "Has Linked Test Data" in the Sources view. If you do not want to automatically split out test data, create new sources and turn off the option when creating the source.

Optional Steps

• Create additional sources. A new source allows you separating a set of data from another set of data, each defined by a unique set of data collection parameters.

a) When creating a source, you need to choose the collection mode. There are 4 possible collection modes: Train Data for Recognition, Train Data for Background, Test Data for Recognition, Test Data for Background. Read more under Collection Modes

b) When creating a source, you have to decide whether you want to automatically split out Linked Test Data or not. You cannot change that parameter after the creation of the source.

c) When creating a source, you have to decide whether you want to surface the source as "collectable" (which makes the source available in both web browser and mobile apps). You can change that parameter at any later time.

d) When creating a source, you have to decide whether you want to Skip the Annotation Queue or not. You can change the parameter at any later time. While "Skip Annotation Queue" is set to False, all uploaded images will be sent to the Annotation Queue where you will need to annotate the images with bounding boxes before being able to use them for training. Skip Annotation is helpful in cases where the pictures coming in contain much more in frame than just the object you want to recognize. In this scenario, you can draw a bounding box around just the recognizable object (effectively cropping it) to give the model a more specific image to look at.

If at any point you need help from the Passio team, please reach out to us at [e-mail]