Hi,
I want to do pose estimation using a fish-eye camera. I’ve tried using skeleton annotator to do this but got the error:
2025-02-05 08:50:41 [34,228ms] [Error] [omni.graph.core.plugin] /Render/PostProcess/SDGPipeline/Replicator_skeleton_data: [/Render/PostProcess/SDGPipeline] Assertion raised in compute - Projection type fisheyeKannalaBrandtK3 is not currently supported.
Just wanted to say that this is a feature that we are interested in having. Currently as a workaround we have small spheres inside the real mesh and are using bounding_box_2d_loose to get the keypoints. This poses some constraints on where we can place the spheres, and in some images (every 10th or so) the spheres are not seen by the detector and therefore missed. When increasing the size of the spheres they are not missed using this approach but then they are seen in the rgb image which is problematic.
I’ve understood from another post that someone has solved this by capturing the rgb-image every second frame and the poses every second frame but this solution must add computational overhead that is undesired.
So whilst I understand how to work around the issue I would prefer if it was possible to perform the projection. I would however be interested if there was some other workaround that you think would be better than the two I suggested.
Thanks in advance!