Detector BYOD: Evaluating on the Target

Now that you have built your BYOD model, you can use the same scripts as in the tutorial for evaluating the pre-trained model. You will need to install your data on the target device and modify the scripts slightly to provide the dataset path.

Evaluate your BYOD Model on the Target

To evaluate the model on the AGX, follow the steps in Recipe Step Three: Evaluate on the Target Device.

As you follow the instructions for Evaluate on Target, keep in mind that to use our evaluation script for the AGX the images and the annotations need to be in the same folder as shown below:

CODE

myvalidationdata
  |---image1.jpeg
  |---image2.jpeg
  |--- ...
  |---*.json

Where *.json is the annotations for the validation set and the images are the validation images.

When you use the run_eval.sh you will need to replace the --dataset_path with the path to your validation data directory (myvalidationdata in the example above).