Volume selection and edge detection possible improvement #201

favreau · 2026-01-19T16:36:03Z

Only set objID when volume has visible content (segment opacity > 0.1%)

…hen volume has visible content (segment opacity > 0.1%)

tarcila

At a very first glance, this PR seems to bring multiple changes:

Get a tighter ID buffer.
Get the actual volume depth instead the current bounding box depth.
Compute some material ID besides the existing object ID. I am not sure about the actual implementation (computing an ID from the transferfunction result). The resulting ID should be stored in a separate AOV, not the object ID one. Also this code does not seem to be wired in. So I guess this is some left over from earlier experiments.

As the goal is to be able to correctly fill the ID buffer, I'd suggest to keep only the code related to the first point.

I have a work-in-progress branch where I do already have that tighter object ID handling (and depth), so if you can wait a bit for the feature, it can be part of a next batch of changes. I still need to discuss with @jeffamstutz about the validity of such a change from an ANARI compliance perspective, but I guess this should be fine.

Last note for completeness, depth should be part of the required information to compute the right ID in case of overlapping volumes, but VisRTX does not support overlapping volumes anyway.

tarcila · 2026-01-19T19:50:11Z

tsd/src/tsd/ui/imgui/windows/Viewport.h

  float m_depthVisualMinimum{0.f};
  float m_depthVisualMaximum{1.f};
-  float m_edgeThreshold{0.5f};
+  float m_edgeThreshold{0.5f}; // 0.0 = 1px radius, 1.0 = 5px radius


Does not seem to be used elsewhere in this PR. Experiments left-over?

tarcila · 2026-01-19T19:51:18Z

tsd/src/tsd/rendering/pipeline/passes/VisualizeAOVPass.cpp


-        // Check if any neighbor has a different object ID (including
-        // background)
+        // Check if any neighbor has a different object ID


Was the previous comment wrong? The related code doesn't look it changed.

tarcila · 2026-01-19T20:04:33Z

devices/rtx/device/gpu/volumeIntegration.h

+    if (contentDepth >= tfar && segmentContentDepth < hit.localRay.t.upper) {
+      contentDepth = segmentContentDepth;
+      if (segmentMaterialID != ~0u) {
+        materialID = segmentMaterialID;


That one does not seem to be unused. It would need to be dropped and the related code.

tarcila · 2026-01-19T20:36:16Z

devices/rtx/device/gpu/volumeIntegration.h

+        &segmentMaterialID);
+
+    // Composite the segment color
+    const float transmittance = (1.0f - opacity);


Feels like we are trying to reimplement part of the existing logic of detail::rayMarchVolume. I feel most of the changes could be done there instead.

Copilot

Pull request overview

This PR enhances volume selection and edge detection by only setting object IDs for volumes with visible content (segment opacity > 0.1%). The change improves the accuracy of object identification and edge detection in volume rendering by ensuring that transparent or nearly transparent volume segments don't contribute to object selection.

Changes:

Modified volume ray marching to track content-based depth and material ID separately from bounding box intersection depth
Introduced a minimum opacity threshold (0.1%) for setting object IDs to avoid selecting transparent volumes
Added a colorToMaterialID function to generate material IDs from transfer function colors for improved edge detection

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
devices/rtx/device/gpu/volumeIntegration.h	Added material ID tracking, content-based depth calculation, and opacity-based objID assignment logic
tsd/src/tsd/ui/imgui/windows/Viewport.h	Added clarifying comment for edge threshold parameter
tsd/src/tsd/rendering/pipeline/passes/VisualizeAOVPass.cpp	Simplified comment about edge detection

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-01-19T21:16:41Z

devices/rtx/device/gpu/volumeIntegration.h

+    // Set objID based on first visible sample (before full accumulation)
+    // Use a very low threshold to ensure consistent objID
+    constexpr float MIN_SEGMENT_OPACITY = 0.001f;
+    if (!objIDSet && segmentOpacity > MIN_SEGMENT_OPACITY && volumeID != ~0u) {


The condition checks volumeID != ~0u but volumeID is initialized to ~0 (which defaults to int type) on line 342. This comparison should use ~0 instead of ~0u for type consistency, or volumeID should be explicitly initialized as ~0u.

Suggested change

if (!objIDSet && segmentOpacity > MIN_SEGMENT_OPACITY && volumeID != ~0u) {

if (!objIDSet && segmentOpacity > MIN_SEGMENT_OPACITY && volumeID != ~0) {

Copilot · 2026-01-19T21:16:41Z

devices/rtx/device/gpu/volumeIntegration.h

    interval.lower +=
        curand_uniform(&ss.rs) * (interval.upper - interval.lower);

  constexpr float OPACITY_THRESHOLD = 0.99f;


The DEPTH_OPACITY_THRESHOLD (0.01f = 1%) differs from MIN_SEGMENT_OPACITY (0.001f = 0.1%) used for objID assignment on line 392. These serve similar purposes (determining visibility) but use different thresholds. Consider consolidating to a single named constant or documenting why different thresholds are needed.

Suggested change

constexpr float OPACITY_THRESHOLD = 0.99f;

constexpr float OPACITY_THRESHOLD = 0.99f;

// Note: This depth threshold (1% opacity) is intentionally higher than the

// minimum segment opacity used elsewhere for objID assignment (0.1%). For

// depth/material reporting we only consider the volume "hit" once it has a

// clearly visible contribution along the ray, to avoid noisy, near-transparent

// samples influencing the first-hit depth and material ID.

Copilot · 2026-01-19T21:16:41Z

devices/rtx/device/gpu/volumeIntegration.h

  }
  return retval;
 }



The colorToMaterialID function lacks documentation explaining its purpose, parameters, and return value. Given its role in material identification for edge detection, a docstring would clarify why RGB quantization is used and how the numBins parameter affects the result.

Suggested change

// Map an RGB color to a discrete material identifier.

// The input color is assumed to be in [0, 1] per channel and is first

// clamped to that range. Each channel is then quantized into @p numBins

// uniformly spaced bins, and the three quantized components are packed

// into a single 32-bit material ID as:

// ID = r * numBins^2 + g * numBins + b

//

// This provides up to numBins^3 distinct material IDs and is typically

// used to assign stable, discrete IDs for operations such as edge or

// material detection based on color.

//

// @param color RGB color in linear space, each component typically in [0, 1].

// @param numBins Number of quantization bins per channel; higher values

// increase color resolution and the number of possible IDs.

// @return A packed material ID derived from the quantized RGB color.

Fix transparent volume selection and edge detection. Only set objID w…

1975c07

…hen volume has visible content (segment opacity > 0.1%)

favreau requested a review from jeffamstutz January 19, 2026 16:37

tarcila self-requested a review January 19, 2026 20:37

tarcila requested changes Jan 19, 2026

View reviewed changes

tarcila requested a review from Copilot January 19, 2026 21:15

Copilot AI reviewed Jan 19, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Volume selection and edge detection possible improvement #201

Volume selection and edge detection possible improvement #201

favreau commented Jan 19, 2026

Uh oh!

tarcila left a comment

Uh oh!

tarcila Jan 19, 2026

Uh oh!

tarcila Jan 19, 2026

Uh oh!

tarcila Jan 19, 2026

Uh oh!

tarcila Jan 19, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Jan 19, 2026

Uh oh!

Copilot AI Jan 19, 2026

Uh oh!

Copilot AI Jan 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	if (!objIDSet && segmentOpacity > MIN_SEGMENT_OPACITY && volumeID != ~0u) {
	if (!objIDSet && segmentOpacity > MIN_SEGMENT_OPACITY && volumeID != ~0) {

-  constexpr float OPACITY_THRESHOLD = 0.99f;
+  constexpr float OPACITY_THRESHOLD = 0.99f;
+  // Note: This depth threshold (1% opacity) is intentionally higher than the
+  // minimum segment opacity used elsewhere for objID assignment (0.1%). For
+  // depth/material reporting we only consider the volume "hit" once it has a
+  // clearly visible contribution along the ray, to avoid noisy, near-transparent
+  // samples influencing the first-hit depth and material ID.

+// Map an RGB color to a discrete material identifier.
+// The input color is assumed to be in [0, 1] per channel and is first
+// clamped to that range. Each channel is then quantized into @p numBins
+// uniformly spaced bins, and the three quantized components are packed
+// into a single 32-bit material ID as:
+//   ID = r * numBins^2 + g * numBins + b
+//
+// This provides up to numBins^3 distinct material IDs and is typically
+// used to assign stable, discrete IDs for operations such as edge or
+// material detection based on color.
+//
+// @param color   RGB color in linear space, each component typically in [0, 1].
+// @param numBins Number of quantization bins per channel; higher values
+//                increase color resolution and the number of possible IDs.
+// @return        A packed material ID derived from the quantized RGB color.

Volume selection and edge detection possible improvement #201

Are you sure you want to change the base?

Volume selection and edge detection possible improvement #201

Conversation

favreau commented Jan 19, 2026

Uh oh!

tarcila left a comment

Choose a reason for hiding this comment

Uh oh!

tarcila Jan 19, 2026

Choose a reason for hiding this comment

Uh oh!

tarcila Jan 19, 2026

Choose a reason for hiding this comment

Uh oh!

tarcila Jan 19, 2026

Choose a reason for hiding this comment

Uh oh!

tarcila Jan 19, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Jan 19, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 19, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 19, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants