search

navis-org / navis-flybrains

Transforms for navis that enable mapping between different Drosophila template brains.
GNU General Public License v3.0
7 stars 2 forks source link

The registration from hemibrain to FAFB is squished #2

Open SridharJagannathan opened 3 years ago

SridharJagannathan commented 3 years ago

Hi, When I try to download a mesh or a skeleton to transform to FAFB like below:

neuron_mesh = neu.fetch_mesh_neuron(1324365879, lod=2, with_synapses=False)
flywireneuron_mesh=navis.xform_brain(neuron_mesh,source='JRCFIB2018Fraw', target='FAFB',verbose=True)

The actual hemibrain mesh looks like below:

Screenshot 2021-03-19 at 14 19 48

However, the transformed stuff looks like squished like below: Screenshot 2021-03-19 at 14 19 14

schlegelp commented 3 years ago

The overall transform looks good as you can see from this co-visualization with the FAFB neuropil mesh.

Screenshot 2021-03-20 at 15 12 08

The transform path is this

1. JRCFIB2018Fraw (voxel) to JRCFIB2018F (nm)
2. JRCFIB2018F (nm) to JRCFIB2018Fum (um)
3. JRCFIB2018Fum (um) to JRC2018F (um)
4. JRC2018F (um) to FAFB14um (um) 
5. FAFB14um (um) to FAFB14 (nm)

This (local) squishing seems to be introduced in step 3. I checked if it had anything to do with my implementation of the Saalfeld lab's Hdf5 transforms but I also tried using the original Java implementation via nat and that produced the same results. I guess we could ask John Bogovic - maybe he'd be able to improve the transform? @jefferis any thoughts?