Output raw video + video showing detections with bounding boxes

I’m looking to capture video output for a proof of concept in hour increments with the intent of:

a) Utilizing the raw video as a source for further augmentation of training data.

b) Randomly sample video with detections for accuracy of the deployed solution.

This video will be uploaded to a cloud provider on a periodic basis or on demand. Can a deepstream pipeline be configured to “tee” the video in this manner? Any other thoughts about how to best approach this?

We have sample applications of uploading data to kafka server. Please install DeepStream SDK through sdkmanager and check