Using TLT to do inferencing on test images, on a FRCNN model, results with kitti output annotations without information on the detection confidence

Using TLT to do inferencing on test images, results with kitti output labels that only contain the class name and the bounding box coordinates, without having information on the detection confidence. This can hugely affect the way we interpret the visualized results of the model.

Refer to [urgent] How to save the bbox confidence probability (i.e. prediction score) during faster-rcnn inference?