Hello GIFTI Team,

I have started updating Caret and the GIFTI Format Document for GIFTI Label Colors and have a few questions.

If a Label is missing color components, how do we color the nodes for this label? Do we use a default color? If so, what color? Or, do we just not apply any coloring to the nodes? My preference is that a default color is used so that the user can see that there is a label associated with a node.

How do we identify nodes that are NOT to be colored? For example, a cortical partitioning scheme may only identify part of the cortex, such as visual areas. In Caret, we use the name "???" to identify nodes that do not receive any coloring. I believe the AFNI palette file uses "NONE" for this purpose. I suppose any label could be used with an alpha of zero. My preference is that we have a name (???, UNIDENTIFIED, NONE) that indicates these unidentified nodes.

John Harwell

Hi John,

Rick and I have discussed some of the defaults that are now in gifticlib but I do not recall what they were. He is Taiwan now with Bob and Gang Chen at an AFNI bootcamp. This makes me quite busy so I may not get around to looking at the code for a little while. In any case, we are quite flexible with regards to the defaults.

cheers,
ziad

Hi John,

In my "gifticlib 1.0.4: Label RGBA attributes" posting, I mention
that if one label has RGBA values, they all must. So as of Oct 27,
the gifticlib implementation requires either a complete color RGBA
listing or none at all.

Personally, I would rather not even allow ambiguity in the color
table. If we are going to forcibly attach colors to the data,
declaring the appropriate way to color the image, then I think the
color table should be complete.

---

A default color does not seem so reasonable to me. If the writing
program wants a default color, it should just write it to the table.
If the writing program is deciding on the coloring, there is no
reason why it should write it in an ambiguous way.

Also, that would remove one basic color from the color map. For
example, if we choose some sort of green as the default, then any
color close to that would make a user wonder whether that node/ROI
has the default color. It would effectively remove variants on
green from the color map.

---

A label that is not to be colored could be written with alpha=0, as
you noted. It makes more sense to me to define what is uncolored
based on the color value, rather than the label text. Of course,
either works.

---

A similar question is what to do for a node value that is not listed
in the label table. It would certainly make sense to not color it at
all. Maybe that should be considered an error case, too. Again, it
would be an incomplete colormap.

I guess my point is that if we are going to allow the writing program
to tell us how to color the image, there should be no ambiguity.

- rick

I think Rick has made some good suggestions.

I agree with the idea of specifying all or none of the color labels, and not allowing partial ambiguous data.
I think it is a good idea to use a 0.0 alpha to represent no coloring.
We sometimes want to dynamic alter the transparency of different structures.

We also agree with Rick's proposal.