DCC production guide
Photogrammetry Point Cloud Reconstruction Guide
Direct answer: Photogrammetry review should treat camera alignment, point cloud quality, mesh reconstruction, normal estimation, and texture reconstruction as separate gates. A model can pass one stage and still fail the next for production use.
Dense Point CloudMesh ReconstructionNormal EstimationPhotogrammetryPoint CloudScan Mesh
Start with camera alignment
Sparse Point Cloud quality tells you whether the photo set aligned consistently. If camera solve quality is weak, later Dense Point Cloud detail only amplifies wrong assumptions. Review overlap, focus consistency, and reflective or textureless surfaces before trying to repair the mesh stage.
Dense cloud to mesh
Dense Point Cloud generation captures more surface detail but also more noise. Mesh Reconstruction must decide what becomes structure and what is trash data. Floating fragments, soft edges, holes, and merged thin parts should be cleaned before the Scan Mesh is treated as final geometry.
Normals and shading
Normal Estimation matters because scan shading can look plausible in one viewer and fail in the next. Recalculate or smooth carefully; over-smoothing destroys carved detail, while noisy normals exaggerate surface speckle.
Texture reconstruction review
Texture Reconstruction is a separate problem from geometry. Exposure mismatch, projection seams, and occluded areas can create blurry or smeared regions on an otherwise acceptable mesh. Review the texture at production distance and under lighting, not only in the reconstruction tool.
Worked example
A scanned statue may have strong camera alignment and a usable dense cloud but still need manual cleanup around fingers, pedestal edges, and shadowed recesses. Approval should include sparse cloud preview, dense cloud cleanup note, reconstructed mesh view, texture pass, and the final decimated game-ready result.
Multilingual production terms
This wording helps a scan review stay precise when alignment, cloud cleanup, meshing, normals, and texture projection fail at different stages.
| 中文 | English | 日本語 | Workflow usage note |
|---|---|---|---|
| 稠密点云 | Dense Point Cloud | 密な点群 | High-resolution surface sample set; contains noise and floating data. |
| 网格重建 | Mesh Reconstruction | メッシュ再構築 | Conversion of point data into triangles; clean before decimating. |
| 法线估计 | Normal Estimation | 法線推定 | Per-point or per-face direction computed from local sampling. |
| 摄影测量 | Photogrammetry | フォトグラメトリ | Image-based reconstruction pipeline with separate alignment, cloud, mesh, and texture stages. |
| 点云 | Point Cloud | 点群 | Set of 3D samples without topology; bridge between photos and mesh. |
| 扫描网格 | Scan Mesh | スキャンメッシュ | Final triangle geometry after reconstruction and cleanup. |
| 稀疏点云 | Sparse Point Cloud | 疎な点群 | Camera alignment result; weak alignment dooms later stages. |
| 纹理重建 | Texture Reconstruction | テクスチャ再構築 | Projected image data on the reconstructed mesh; review exposure and seam problems. |
Related pages and next step
Open SEELE AI workspaceFAQ
Why is the sparse cloud still worth reviewing after a dense cloud exists?
Because bad alignment at the sparse stage often explains downstream holes, warped geometry, and poor texture reprojection.
Can a scan mesh be approved before texture reconstruction is fixed?
Sometimes for geometry-only use, but not when the delivery depends on reconstructed texture quality.
What is a common normal-estimation mistake?
Smoothing away sharp carved forms or leaving noisy micro-normal chatter that survives into lighting and baking.
What is strong photogrammetry proof?
A sequence showing alignment quality, dense cloud cleanup, mesh reconstruction, texture result, and the final cleaned asset in the target runtime or DCC.