Kitti confidence values in inference labels

• AWS
• YoloV4 TAO

When inference on a trained model a kitti format label file is generate. It has the confidence values included at the end of each object detection. Is there a way to not add the confidence value to the kitti label file? It’s preventing me from importing the file into cvat to evaluate and correct the bounding boxes.

I suggest you to run “cut” or “awk” or something else to deal with the confidence value.

Thanks for the fast answer,

I can do it many ways, excel, programmatically, manually. I was hoping there was access to the python script that adds the confidence value and remove it.

I would like to use it to pre label the data, import it into a labelling tool such as cvat or labelmg, etc correct the errors and use the corrected data to train again in TAO.

There are many steps. I’m trying to reduce them as much as possible.

Any ideas? I have tested Onepanel. This entails training another model as it doesn’t support .TLT/ETLT models. So it’s doubling the process further.

I’m looking forward to the UI version of TAO being released to be able to speed up the entire process.

So, do you mean you want to use one network to run inference in order to get some pre label files? You can run triton-app to do yolov3 inference. Please follow steps in GitHub - NVIDIA-AI-IOT/tao-toolkit-triton-apps: Sample app code for deploying TAO Toolkit trained models to Triton .
Need to modify tao-toolkit-triton-apps/yolov3_postprocessor.py at main · NVIDIA-AI-IOT/tao-toolkit-triton-apps · GitHub to delete the confidence value.

Thanks I will take a look

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