Omniverse Realtime Viewer

Mesh X-Ray Delivery Profile

Use this focused reference when the requested X-Ray/technical look must work on ordinary UsdGeomMesh geometry: tessellated props, support slabs, fixtures, or converted CAD meshes. It complements the canonical presentation recipes; it does not replace a BREP-aware adapter when one is available, and it introduces no modeling-kernel dependency.

Select The Geometry Provider First

Inventory the source before choosing a line implementation. If an exact BREP-aware provider is present and the task requires its authored topology, use that provider. If the stage is UsdGeomMesh-only or no BREP provider is available, use this profile. For a hybrid stage, select a provider per prim and compose their segments through the same camera/presentation contract.

Do not infer this from the user’s word “CAD”: converted CAD is often ordinary mesh geometry. Record the selected provider and the count of included/excluded meshes in the effective style capability response.

What Mesh X-Ray Can And Cannot Show

DepthSD and the standard color/normal/ID AOVs describe the foremost rasterized surface. They cannot reveal arbitrary layers behind it. Treat them as a surface-shading and visible-edge signal, not as hidden-line data or a substitute for depth peeling.

To reveal topology through regular meshes, extract deterministic mesh segments, then rasterize those segments against a private current-camera depth test. A light shell makes the scene read as X-Ray; the topology pass carries the through-geometry information. One depth test can classify front/rear segments, but cannot truthfully reconstruct every deeper overlapping layer. Keep rear strokes quieter, or use a deliberate all-feature continuation convention; do not imply physical transmission or multilayer depth peeling.

Recommended Mesh Render Graph

LdrColor + NormalSD + InstanceSegmentationSD (+ enhanced DepthSD)
    -> SPG shell and visible-guide node(s)
    -> SPG current-camera mesh projection/private-depth/stroke nodes
       -> visible graphite features
       -> subdued hidden/continuation graphite features
    -> published RGBA output -> existing delivery-format conversion

Validate guide semantics, then wire them into the same-frame SPG graph. Keep the shell, topology raster and RGBA composite on GPU. Map only the processed output for presentation. An explicitly retained legacy GL readback/upload path needs separate capability and timing evidence; caching an old camera overlay does not establish current-camera SPG delivery.

AOV Capability Tiers

The canonical X-Ray recipe does not require DepthSD. The enhanced mesh tier does, because its local-depth edge formula consumes it. Never silently enable a depth-derived edge from an unvalidated or mismatched tensor.

Mesh Topology Rules

Start from the LineGeometry adapter in line-compositor-delivery.md.

• Include every render-visible mesh in the style scope, including passive tables, plates, supports, and occluders. Do not limit line extraction to selectable, animated, or currently inspected prims.

• Keep boundary and sufficiently noncoplanar crease edges; reject coplanar tessellation diagonals. Preserve silhouette candidates separately.

• Sort and cache topology by mesh/transform revision. Reproject and rasterize it from the newest applied camera pose whenever that pose changes.

• Use the exact OVRTX viewport, camera transform, projection, depth convention, and top-left image origin. A line layer that is even one camera update late reads as outline latency during orbit.

• Prefer a same-device topology pass for dense meshes. If a GL-to-host-to-CUDA fallback misses the current-camera refresh budget, report deferred lines rather than presenting stale lines as live X-Ray.

Visible and hidden strokes may differ in value or dash pattern, but keep them in one color family. A useful noncanonical graphite family is visible RGB (0.215, 0.240, 0.267) with a modestly lighter, lower-alpha hidden variant. This preserves depth readability without a competing cyan line color.

Accurate Depth Versus Fast Continuation

Use the private-depth line pass in line-compositor-delivery.md when a style claims visible/hidden depth classification. It rasterizes occlusion triangles and is the portable accuracy profile.

Use a same-device continuation profile when the product prioritizes responsive technical readability over exact hidden-line classification. Keep persistent GPU arrays for world vertices [V,3], edge vertex indices [E,2], adjacent face indices [E,2], face normals/centers, projected vertices, validity flags, and two 4x4 camera matrices. Per frame:

1. Copy only the committed view/projection matrices into persistent GPU arrays.

2. Project vertices, rejecting nonfinite points and clipping segments before the perspective divide.

3. Classify an edge from adjacent face/view signs as silhouette, front feature, rear continuation, or reject.

4. Raster visible marks then hidden/continuation marks into the app-owned BGRA buffer with a deterministic priority rule: visible wins over hidden and either wins over the shell.

5. Bound samples by the clipped projected length and a configured maximum; log and skip any over-budget span instead of creating unbounded work.

This direct GPU profile is not a private-depth substitute. Label it continuation, not depth_classified_hidden, and retain a per-style switch to the accurate depth path when the distinction matters.

Explicit Style Scope

Use a viewer-owned scope predicate, evaluated after inherited visibility and purpose. Include default/render UsdGeomMesh prims and passive supports by default. Exclude lights, cameras, helpers, guide geometry, and environment or backdrop meshes only through named application configuration or inherited opt-out metadata. Never exclude an object merely because it is not selected, animated, or editable. Recompute the scope on stage/visibility/purpose changes and report the resulting mesh/edge counts.

Surface-Forward Shell

Do not turn geometry-only mode into a white or uniform-blue fill. That makes real mesh geometry read as pure outlines and removes the form cues that make the X-Ray useful. Suppress only the shell coverage of a large foreground occluder when necessary; retain normal-shaped blue fill on every surface.

For the optional mesh surface-forward profile, let rim = smooth((1 - abs(Nz) - 0.08) / 0.72) and facing = smooth((Nz + 1) * 0.5). A proven display-referred shell is:

R = 0.39 + 0.17*facing + 0.05*rim
G = 0.63 + 0.22*facing + 0.06*rim
B = 0.76 + 0.19*facing + 0.07*rim

Composite visible guide edges and explicit topology only after this shell. The formula is a mesh delivery profile, not a change to the canonical X-Ray palette in style-recipes.md.

Guide Cleanup

Use ID boundaries, normal creases, and relative local depth discontinuity only as visible-surface accents. Suppress guide pixels with ID 0; do not promote border-connected background or flat slab artifacts into object lines. Gate fine depth contours by local normal/curvature evidence, and reject isolated screen-space fragments that cannot be associated with a mesh feature or a coherent silhouette. This prevents disconnected dark marks from becoming more prominent than the topology drawing.

Do not infer hidden topology from a depth contour. When a contour is needed as a stylistic accent, use it at a lower weight than the mesh line pass and test it under camera motion.

Interaction, Cost, And Proof

Measure an X-Ray-off and X-Ray-on orbit separately. Report server frame time, stream submission, browser decode/presentation, and the current-camera line refresh rate. Attribute cost to guide mapping, shell, topology raster, and any host transfer independently; a fast renderer with a stale line layer is not a real-time X-Ray result.

Capture these mesh-specific proofs in addition to the standard style matrix:

1. An above-support three-quarter frame showing shaded cube/sphere/cone-like forms, not only outlines.

2. A lateral orbit that begins from the fitted above-support camera.

3. A below/behind-occluder frame where a passive support edge remains visible through the foreground geometry.

4. A fixed-camera comparison with the shell enabled and with the surface-forward profile enabled, confirming that the latter retains form.

Reset or fit the camera explicitly before capturing the above-support orbit. Reconnecting a streaming client does not guarantee a fresh camera state. Record the camera generation used by the shell and topology pass; they must match for every frame in the proof.

Label this result presentation_xray_mesh. It is a technical-illustration composite over regular mesh geometry, not physical transparency and not a claim of arbitrary-layer visibility.

Full wireframe and feature cleanup

Keep smooth surfaces free of accidental tessellation spokes in feature mode. Before adjacency counting, remove faces collapsed by welding, and suppress duplicate-index smooth seams and numerical micro-segments. A nonmanifold condition must be reported rather than assumed to express design intent.

Offer full triangle wireframe only as an explicit diagnostic mode: include all accepted triangle edges and diagonals, declare whether rear edges remain visible, and keep its optional no-fill background distinct from the X-Ray shell. Verify coverage numerically and ensure line disable removes marks. See the SPG resource and geometry contracts in line-compositor-delivery.md.