Hi, I’m working through the Replicator “Online Generation” example but I’m getting an error because the “boundingBox2DTight” groundtruth value is empty:
example:
https://docs.omniverse.nvidia.com/app_isaacsim/app_isaacsim/tutorial_replicator_online_generation.html
This line generates an error (because of empty gt_bbox):
bboxes = torch.tensor(gt_bbox[[“x_min”, “y_min”, “x_max”, “y_max”]].tolist())
I’ve output the “rgb” groundtruth to an image and it looks ok to me (i.e. there should be bounding boxes).
Trying to figure out what’s wrong but any help would be appreciated
Thanks
Correction, the failure line is:
areas = (bboxes[:, 2] - bboxes[:, 0]) * (bboxes[:, 3] - bboxes[:, 1])
because bboxes tensor is “tensor()”
Here is an example of the “rgb” groundtruth which results in empty bounding boxes:
I’ve also tried
verify_sensor_init=True
wait_for_sensor_data=2
in self.sd_helper.get_groundtruth
For anyone stumbling onto this - I have a fix.
I added:
sd_helper.initialize(sensor_names=sensor_names, viewport=viewport)
Which is what is done in
standalone_examples/replicator/visualize_groundtruth.py
And that solved the issue for me.
2 Likes