Need to process individual shape features and merge them again according to batched merged octrees

[gargsid]

I understand the octrees are merged in an input mini-batch. I need to process some features of individual shapes and then again combine the processed features so that they can are merged again.

For example, I am trying this

print('seg_header_pts feature:', feature.get_shape())  # output features of the HRNet backbone (1, 928, ?, 1)
index = octree_property(octree, property_name='index', depth=depth_out, dtype=tf.float32, channel=928) # (928, ?)

Is the argument channel=928 correct in the above code? And how do I get the per-point features of a single shape now?

Furthermore, how do I merge all the single per-point features according to the merged features again?