search

BIC-MNI / Display

MNI Display project, converted from CVS
Other
8 stars 4 forks source link

RGB images #49

Open ajwood opened 8 years ago

ajwood commented 8 years ago

Display used to be able to open RGB volumes, if the fastest-varying vector_dimension contains RGB triples (on the range 0-1), but this doesn't seem to work with the latest version.

Is there any change that this feature could come back? Or alternatively, is there a different way to open RGB volumes in Display?

rdvincent commented 8 years ago

It wasn't taken out on purpose - I'll figure out what happened and fix it.

rdvincent commented 8 years ago

@ajwood Should be fixed if you check out the latest libminc and bicpl. I also make a minor change to Display that formats the RGB value of the current voxel.

ajwood commented 8 years ago

Looks great! I'm wondering, do you think it'd be tricky to get trilinear or tricubic interpolation working?

rdvincent commented 8 years ago

@ajwood I assume you mean trilinear/tricubic with RGB. It's probably not that tricky. How useful would it actually be?

ajwood commented 8 years ago

@rdvincent Right; I mean trilinear/tricubic interpolation for RGB slices (from the menu: volume config (Q) -> interp (S)). Currently it shows the interpolant changing, but doesn't actually affect the display.

Do you think it'd be less useful for RGB than it is for intensity images?

rdvincent commented 8 years ago

@ajwood I don't really have an opinion - I'm just trying to get a general sense of the priority.

ajwood commented 8 years ago

@rdvincent My main uses are this:

Digital microscopy: Interpolation mainly useful for zooming and collecting snapshots.

DTI colour maps: Interpolation helpful when the maps are noisy.

I also wonder about visualizing deformation fields coming out of minctracc (or other registration tools) in Display. Not sure what this would look like, or how it'd play with the different interpolants.

Other people might have different thoughts.

rdvincent commented 8 years ago

@ajwood I'll look into it in any case. Do you have any appropriate images available that I can use for testing.

ajwood commented 8 years ago

Sure, here's a dti map.

gdevenyi commented 8 years ago

I have a related request in #15 to be able to do additive RGB with multiple volumes, but in that case I'm hoping for independent adjustment of the colourmaps for each volume.

In general the idea here is to experiment with "hybrid contrast" for our segmenters, in the hopes of improving tissue differentiation.

rdvincent commented 8 years ago

@ajwood I recently added trilinear interpolation for RGB. Tricubic is harder because of implementation details in the volume_io library.

ajwood commented 8 years ago

I just noticed that the segmentation painting threshold doesn't work for RGB data. Would there be an easy fix for this? e.g. threshold on the mean intensity