DCC production checklist
Realtime Rendering Visibility Checklist
Direct answer: Approve realtime visibility only after the target camera path proves LOD Screen Size thresholds, frustum and occlusion behavior, masked cutoff stability, translucent sorting, and Screen Space Reflections at the project’s minimum resolution and performance profile.
Screen SizeFrustum CullingMasked ClippingOcclusion CullingScreen Space Reflections
1. Measure projected coverage before choosing LODs
Record each LOD transition as projected Screen Size, because field of view and resolution change the result. A prop baseline is LOD0 above 0.35 screen fraction, LOD1 from 0.35 to 0.15, LOD2 from 0.15 to 0.05, then a billboard or Cull Distance. Treat these as starting thresholds. Scrub the fastest camera and reject transitions that alter silhouette by more than two pixels.
2. Prove frustum and distance culling bounds
Frustum Culling depends on bounds, so visualize bounds after animation, vertex displacement, and world-position offset. Tight static bounds may clip moving foliage or VFX before the geometry leaves frame; oversized bounds keep objects rendered and weaken Occlusion Culling. Set Cull Distance from gameplay importance and projected pixels, not merely object class. Test the screen corners, ultrawide aspect ratio, and camera cuts. A one-frame disappearance at an edge is evidence of incorrect bounds, not an acceptable optimization artifact.
3. Calibrate occlusion without causing popping
Use occlusion visualization to confirm large opaque occluders reject hidden objects without making small movable props into costly occluders. Queries or hierarchical Z-buffer tests can lag during rapid motion, so run sprint turns and camera cuts. If an object pops, expand bounds only for measured motion or adjust occlusion tolerance; disabling Occlusion Culling globally hides the fault and forfeits the budget. Record visible primitives and GPU time.
4. Choose masked or translucent from the material need
Use Masked Clipping for binary cutouts such as leaves, chain-link, and torn cloth when hard coverage is acceptable. Start opacity cutoff near 0.33–0.5, then inspect mip levels; a fixed cutoff can thin distant foliage, so coverage-preserving alpha mips may be required. Use Translucent Blending only for graded transmission such as glass or smoke. It incurs overdraw, sorting limits, and often reduced depth participation. Never choose translucent merely to soften a poor alpha edge.
5. Inspect SSR failure signatures at screen edges
Screen Space Reflections can reflect only visible screen data. Typical symptoms are reflections disappearing at the viewport edge, stretching behind foreground objects, leaking across depth discontinuities, and flickering on thin rough surfaces. Test at the minimum dynamic-resolution scale, because fewer samples amplify gaps. Increase roughness or use reflection captures, planar reflections, or hardware ray tracing according to importance; raising SSR quality cannot reconstruct off-screen information. Capture a camera pan, not a flattering still image.
6. Apply one quantified visibility decision
For a waist-high lobby plant, begin with LOD thresholds 0.30, 0.12, and 0.04, a 6,000-unit Cull Distance, masked leaves at 0.4 cutoff, and no SSR on foliage. A test reveals two-pixel leaf loss at LOD1 and one-frame edge clipping during wind. Keep the cull distance, move LOD1 to 0.10, enable alpha-coverage mips, and expand bounds by measured wind displacement. Accept only when the same path holds 60 fps and shows no visible pop at 1080p.
7. Read symptoms before changing quality knobs
Geometry vanishing only at frame edges indicates bounds or Frustum Culling; disappearance behind a wall followed by delayed return indicates occlusion latency. Sparkling leaf silhouettes indicate masked mip coverage or temporal antialiasing, while glass layers reversing order indicate translucent sorting. Reflections that vanish with the source object off-screen are normal SSR limitations, not bad UVs. Change one control at a time and replay the same camera path so a quality increase does not accidentally conceal a separate visibility defect.
8. Record reproducible visibility evidence
State engine build, map, camera FOV, resolution scale, anti-aliasing, LOD Screen Size values, bounds, Cull Distance, masked cutoff, translucency sorting, and reflection method. Include LOD-color captures, bounds visualization, occlusion statistics, overdraw, GPU profile, and a real-time camera video. Use the terms below in multilingual review notes.
| zh | en | ja | Production context |
|---|---|---|---|
| 屏幕尺寸 | Screen Size | スクリーンサイズ | Projected coverage and LOD transition |
| 视锥剔除 | Frustum Culling | 視錐台カリング | Camera-frustum visibility and bounds |
| 遮罩裁剪 | Masked Clipping | マスククリップ | Binary opacity cutoff and mip coverage |
| 遮挡剔除 | Occlusion Culling | オクルージョンカリング | Hidden-object rejection and latency |
| 屏幕空间反射 | Screen Space Reflections | スクリーンスペース反射 | Screen-data reflection limitations |
| 透明混合 | Translucent Blending | 透明ブレンド | Graded opacity, overdraw, and sorting |
Realtime Rendering Visibility Checklist FAQ
What causes a realtime mesh to disappear at the edge of the camera?
First inspect rendered bounds. Animation, wind, or vertex displacement may extend beyond static bounds, causing Frustum Culling before the visible pixels actually leave the frame.
When should foliage use masked instead of translucent shading?
Use masked shading when coverage can be binary. It writes depth more predictably and avoids many sorting costs; preserve distant silhouette with alpha-aware mips and a tested cutoff.
Why do screen-space reflections vanish during a camera pan?
SSR has no off-screen source data. Edge fade and missing reflections are expected limits; use reflection captures, planar reflections, or ray tracing for surfaces that require stable off-screen reflections.