Adding background class to decrease random background detections

Currently TLT models doesn’t supporting the background class as an object. Due to this the trained pipeline is loosing the ability to differentiate foreground object and background. Because of this, the models trained on TLT is showing random backgrounds as objects.

Do you know any way to reduce this behaviour in current pipeline. If yes please let me know. If not then please consider adding this in the upcoming release.

Moving this topic from Maxine forum Maxine - NVIDIA Developer Forums into TAO forum.

May I know which network do you focus on?

Yolo_v3 and Yolo_v4

Refer to Training with Adversarial/False Positive Data - #3 by Morganh

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.