Get data from Annotator with multiple cameras

Omniverse Isaac Sim
1

I’m not able to get rgb image from the annotator with multiple cameras.
So far i;ve create the cameras/render_product/annotator in the following way as the examples were suggesting.

       self.render_list = []
        for i in range(self._num_envs):
            camera = rep.create.camera(
                position = (self.ws_cx, self.ws_cy , 1.5),
                rotation = (0,-90,0),
                focal_length = 50,
                horizontal_aperture = 90,
                clipping_range = (0.1, 1.5+0.5),
                parent = "/World/envs/env_"+str(i),
            )
            self.render_list.append(rep.create.render_product(camera, (640, 480)))

        self.rgb = rep.AnnotatorRegistry.get_annotator("rgb")
        self.rgb.attach(self.render_list)
        print('annotator attached')

However, when i call self.rgb.get_data(), i only obtaining a single image of dimension 640,480 while i was expecting the numpy array to have a different dimension.

3

Hi @user126140 - Someone from our team will review and respond to your question.

4

Hi there, here is similar example using multiple camera with writers and annotators.

In the current version one needs to create an annotator for each render product. In the example we have a note about this:

# Create render products
rp1 = rep.create.render_product(str(camera_prim1.GetPrimPath()), resolution=(320, 320))
rp2 = rep.create.render_product(str(camera_prim2.GetPrimPath()), resolution=(640, 640))
rp3 = rep.create.render_product("/OmniverseKit_Persp", (1024, 1024))

# Acess the data through annotators
rgb_annotators = []
for rp in [rp1, rp2, rp3]:
    rgb = rep.AnnotatorRegistry.get_annotator("rgb")
    rgb.attach([rp])
    rgb_annotators.append(rgb)

# NOTE A list of render products will be supported in the near future, currently only the first render product in the list will be used
# rgb = rep.AnnotatorRegistry.get_annotator("rgb")
# rgb.attach([rp1, rp2, rp3])

Let me know if you need further assistance.

Best,
Andrei

5

@ahaidu Thank you for the reply, i was also wondering if there is the possibility to stop the annotator to get images and depth information. In the application I’m working on, I just need an image at a certain moment.

Should I create a render_product each time I need the observation of the scene? Is there the possibility to stop the annotator? I have’t found an answer in the documentation

6

  • you can use annotator.detach() once the annotator is no longer needed

  • if the camera and the resolution stays the same, you can (and should – since every new render product will use up the GPU load) re-use the existing render product

  • currently (in version 2022.2.1) any created render product cannot be destroyed on the fly and will add to the gpu load, this should be fixed in the new release. Creating a new stage will however destroy them and release the free the gpu load.

7

@user126140 here is a new, more comprehensive get_data example in our docs:
https://docs.omniverse.nvidia.com/app_isaacsim/app_isaacsim/reference_python_snippets.html#get-synthetic-data-at-custom-events-in-simulations

8

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