Collecting our own Detection Datasets with pictures

I have follow the tutorial : Collecting your own Detection Datasets.
And I wonder how can we use picture (without already determined box of detection) or should we use only the device’s video stream of a camera?

May I know which Jetson platform and JetPack version you used?

I use a Jetson NX and the last version of JetPack (4.5.1)


You can use pictures but please create the label file with the rule mentioned below:


If you already have the pictures, you can use CVAT tool to annotate them. Then export them in Pascal VOC format and as Aasta mentioned, create a labels.txt file for it.