feat(segmentation_user_layer): add user shader for segmentation layers with support for segment properties by chrisj · Pull Request #1074 · google/neuroglancer

chrisj · 2026-07-13T16:59:53Z

continuation of #862

seankmartin

This looks nice, thanks for all the changes! In general the API in the shader seems pretty reasonable to me. I've not gotten to go through it all, but some small comments from what I have looked through. There are also some TODO markers and the likes, not sure if those are still relevant or you settled on the approach in those cases

seankmartin · 2026-07-16T10:41:01Z

-  colorGroupState.segmentColorHash.compute(color, objectId);
-  return color;
+  const { getShaderSegmentColor } = displayState;
+  return getShaderSegmentColor(objectId) ?? color;


Could we detect if the shader is the default one, and in that case use the old non-shader color path to display the colors in the UI for performance. I noticed that the segmentation menu can sometimes feel a bit more sluggish to go through, which seems pretty fair when you've chosen to use the segment shader and to me it seems fine. But if possible to avoid this slight hit if you didn't change your shader I think that would be good

I think that is what I should do for now but maybe I should give a quick try at adding batching. I don't notice sluggishness but I do notice that the color doesn't load immediately if scrolling very fast on my machine. I can't replicate that on the main branch.

It's easy to add support to calculate many segment colors in a single draw call, the issue is that the segment list is designed so that each row is processed independently. But if I can get it working, it should be far more performant.

That makes a lot of sense, thanks Chris

@seankmartin can you check if this deployment has noticeably better performance for you?

demo

Oh wow, that's so much better yes thanks @chrisj

seankmartin · 2026-07-16T10:44:16Z

+      const hasStringProperty = inlineProperties.properties.some(
+        (p) => p.type === "string",
+      );
+      if (!hasStringProperty) {


I guess this is leftover but it does raise what we should do here. Shader error comes to mind, but then in theory the underlying data could change and lose a property. Maybe that's fine though.

yeah that was for demonstration as string properties are not common. But do you mean if a shader references a property that was later removed from the segment properties file? If so I think the current error "property does not exist" is correct.

seankmartin · 2026-07-16T10:52:18Z

+  });
+}
+
+function makeSkeletonShaderCodeWidget(layer: SegmentationUserLayer) {


Minor but the style of the element layout seems different on this new shader controls in the UI compared to the skeleton ones

seankmartin · 2026-07-16T10:57:31Z

 }

+export const DEFAULT_FRAGMENT_SEGMENT_COLOR = `
+vec3 segmentColor(vec3 color, bool hasProperties, bool isStated) {


Oh also minor but leading newline in the resulting shader here that probably is better left out

also I should rename it from FRAGMENT becomes it is sometimes used in the vertex shader

chrisj · 2026-07-16T15:46:15Z

@seankmartin thanks for the review! I have some changes I'll push shortly that I noticed yesterday

seankmartin · 2026-07-16T15:53:49Z

@seankmartin thanks for the review! I have some changes I'll push shortly that I noticed yesterday

sounds great, thanks!

chrisj · 2026-08-17T11:09:24Z

@seankmartin I simplified the performance optimizations and added it to this PR. I added a temporary setting that you can toggle with "q" to easily observe the difference. Can you give it a test to see if you are still see the big performance improvement? I can see the a difference when I press "q" on and off while dragging the scroll bar. If it looks good I'll revert those last few commits.

example

seankmartin · 2026-08-17T11:12:28Z

@seankmartin I simplified the performance optimizations and added it to this PR. I added a temporary setting that you can toggle with "q" to easily observe the difference. Can you give it a test to see if you are still see the big performance improvement? I can see the a difference when I press "q" on and off while dragging the scroll bar. If it looks good I'll revert those last few commits.

example

Thanks @chrisj, that's great! And yes still much better for me when batched

…s + segment property ui control + segment properties in invlerp control

segment_list: add prefetchBaseObjectColors - computes colors before rows are created add updateMany - recomputes colors for all currently rendered rows following display-state changes virtual_list: add optional VirtualListSource.prefetch(start, end) callback. It is called before missing rows in that interval are constructed

chrisj · 2026-08-22T01:47:04Z

  toMerged: Uint32Array,
 ): InlineSegmentNumericalProperty {
-  const values = new Float32Array(numMerged);
+  const values = property.values.slice(0, numMerged);


I've been diving into this change and need some thoughts.

This function is called when merging multiple segment property maps together. It was casting every numerical property values array to Float32Array. If the maps had differing sets of segments, each segment received a value of NaN for the numerical property of each map it was not in.

The easiest solution would be to just use a zero value and keep each values as it's original numerical type. Another option is to do that but add a validity mask. This is a rough idea of what that would look like: AllenInstitute@4e7a970

if (hasProp("myProperty")) { uint value = prop("myProperty"); // use value }

I'm leaning towards using 0 instead of NaN for non-float and skipping the validity mask as I imagine it won't have much use.

Seems reasonable to me to do the simpler version, I also can't imagine it's used too often

…segment properties, remove todo comments

Previously, merged numerical properties were initialized with NaN, which coerced integer typed arrays to zero while leaving float properties as NaN. Allocate the merged values with the source typed-array constructor so missing entries consistently default to zero, preserve the property datatype, and expand the bounds to include zero.

chrisj · 2026-08-25T12:19:18Z

 ): InlineSegmentNumericalProperty {
-  const values = new Float32Array(numMerged);
-  values.fill(Number.NaN);
+  const values = new (property.values


@jbms are you good with this change?

759a4d6

chrisj force-pushed the cj-segment-color-user-shader branch 8 times, most recently from 07f7c7c to 7adf3fc Compare July 15, 2026 16:43

chrisj marked this pull request as ready for review July 15, 2026 16:43

chrisj force-pushed the cj-segment-color-user-shader branch from 7adf3fc to 53cfe9b Compare July 15, 2026 16:45

seankmartin reviewed Jul 16, 2026

View reviewed changes

chrisj force-pushed the cj-segment-color-user-shader branch 2 times, most recently from 00444f6 to 9f4e76c Compare July 20, 2026 15:28

chrisj force-pushed the cj-segment-color-user-shader branch 2 times, most recently from 72bcce9 to b3fd516 Compare August 13, 2026 16:56

chrisj force-pushed the cj-segment-color-user-shader branch from 7317500 to b3fd516 Compare August 17, 2026 13:34

chrisj added 6 commits August 18, 2026 16:42

feat(segmentation_user_layer): add user shader for segmentation layer…

93725f6

…s + segment property ui control + segment properties in invlerp control

cleanup

234e312

fix: getBaseObjectColor should not add hover highlight

a4c88c3

cleanup

bb0946e

fix segment color shader ui now properly styles the header

7c1e1bc

chrisj force-pushed the cj-segment-color-user-shader branch 2 times, most recently from fc7512d to 775e56d Compare August 18, 2026 21:45

cleanup

c4c7d9d

chrisj force-pushed the cj-segment-color-user-shader branch from 775e56d to c4c7d9d Compare August 19, 2026 14:55

support fallback shader parameters for segment color shader

f8708b1

chrisj commented Aug 22, 2026

View reviewed changes

chrisj added 8 commits August 24, 2026 16:54

clean up shader property references to either refer to annotation or …

77f332a

…segment properties, remove todo comments

fix(skeleton): apply segment and layer opacity

dd75a15

fix(segmentation): guard fallback highlight uniform

ac0c1c6

fix(segment-color): detect formatted vec3 shaders

d9a47bd

fix(invlerp): handle empty and uint64 property values

e25108e

fix(segmentation): aggregate shader errors by render path

55d341e

fix(segment-color): preserve large string literal ids

1784f69

chrisj force-pushed the cj-segment-color-user-shader branch from 7dac20e to 1784f69 Compare August 25, 2026 12:16

chrisj commented Aug 25, 2026

View reviewed changes

Uh oh!

Conversation

chrisj commented Jul 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

seankmartin left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

chrisj commented Jul 16, 2026

Uh oh!

seankmartin commented Jul 16, 2026

Uh oh!

chrisj commented Aug 17, 2026

Uh oh!

seankmartin commented Aug 17, 2026

Uh oh!

chrisj Aug 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

chrisj commented Jul 13, 2026 •

edited

Loading

chrisj Aug 22, 2026 •

edited

Loading