How to Use Test Set and Ground Truth for MaP Evaluation in DetectNet_v2?

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Objective:
To calculate the Mean Average Precision (MaP) of a detectnet_v2 model using a test set.

Context:
I have a test set consisting of 800 images, along with their ground truth bounding boxes provided in the Kitti format. I want to evaluate the test accuracy of a DetectNet_v2 model.

I noticed in the provided notebooks that there is an option to run inference on test images, which outputs images with predicted bounding boxes and a corresponding label file containing these bounding boxes. Is there a way to use these predicted bounding boxes and their ground truth bounding boxes to calculate the MaP of the model?

3

Yes, there is an option to run evaluation. You can run tao model detectnet_v2 evaluate.

4

Hi @Morganh ,

Thank you for your reply.

Can you please provide any documentation on how to use the evaluate function?
I can see from the notebook that it takes the same specs used for training as input. Where can I provide my validation data path in the specs?
Is there any other configuration that I should keep in mind while using the evaluate function on a custom test set?

5

Suggest you to run default notebook to get familiar with it.
You can generate tfrecords files based on your own dataset.

6

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

7

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