I was wondering if the training of CenterSnap requires ground truth 6D pose for the objects? If so, was it something that directly passed into the encoder-decoder, or it is completely separate and added to the 3D map vectors for each pointcloud? And how did you get the ground truth object pose for the dataset, if that's the case?
I was wondering if the training of CenterSnap requires ground truth 6D pose for the objects? If so, was it something that directly passed into the encoder-decoder, or it is completely separate and added to the 3D map vectors for each pointcloud? And how did you get the ground truth object pose for the dataset, if that's the case?