实验 24:A18 Topology-constrained Pareto Router

把水密拓扑硬约束和输入视角 silhouette IoU 合并成无 GT Pareto 路由

实验定位 混合偏正 / Pareto frontier

A18 把 A15 的 topology candidate routing 和 A17 的输入视角 silhouette IoU 合并:非 baseline 候选必须先通过水密、boundary=0、nonmanifold=0、IoU 不低于 baseline 容忍带等硬约束,再用 Pareto score 在 IoU、水密、面片数、mask 面积漂移之间排序。选择阶段不使用 GT。

结果比 A17 更合理:不再让 silhouette 单独选择非水密候选;同时相对 baseline 获得 Chamfer 略优、F@5 提升、IoU 提升和面片数大幅下降。但由于 2 个 case 回退 baseline,水密率是 80%,说明还需要更细的 per-view 修复/路由。

实验设计(Image-2 风格)

实验设计图:topology-constrained Pareto routing
实验设计图:topology-constrained Pareto routing
模块设计图:view-consistent topology-constrained routing
模块设计图:view-consistent topology-constrained routing

模块设计(Image-2 风格)

Hard Gate

  • baseline always pass as fallback
  • non-baseline candidates must be watertight
  • boundary_edges == 0 and nonmanifold_edges == 0
  • mean silhouette IoU >= baseline IoU - 0.02
  • min silhouette IoU >= 0.05
  • silhouette area ratio in [0.20, 2.20]

Pareto Score

0.60*IoU_delta + 0.30*watertight + 0.18*compactness - 0.14*area_drift - 0.08*face_ratio_drift

choose max score after topology hard gate; GT metrics used only after selection

实验结果(表格)

方法成功Pareto scoreIoUIoU ΔChamferChamfer ΔF@5F@5 Δfacesfaces Δ水密率
baseline10/10-0.05970.4618-0.08161-0.4336-950660-0.0%
ranker_selected10/10-199.59500.46910.00730.081670.000060.43960.0059406873-54378790.0%
repair_hybrid10/10-199.59500.46710.00530.081930.000320.43880.0052443226-507435100.0%
silhouette_routed-/-----------
pareto_routed10/100.19230.47770.01590.08132-0.000290.43840.0048382922-56773980.0%
alpha0259/9-999.00000.46570.00130.083250.000090.42960.00081040042-147480.0%
selected_alpha9/9-999.00000.4608-0.00360.084170.001010.43140.00261048883-59060.0%

每个 case 的路由

case最终选择可选候选scoreIoU ΔChamfer ΔF@5 Δfaces Δ水密率
gso_000_input4ranker_selected['baseline', 'ranker_selected', 'repair_hybrid']0.37110.1400-0.001600.0202-10762100.0%
gso_001_input4ranker_selected['baseline', 'ranker_selected', 'repair_hybrid']0.1455-0.0013-0.001050.00823056100.0%
gso_002_input4ranker_selected['baseline', 'ranker_selected', 'repair_hybrid']0.27680.0228-0.003770.0502-1227506100.0%
gso_003_input4ranker_selected['baseline', 'ranker_selected', 'repair_hybrid']0.2684-0.0001-0.00005-0.0041-2330592100.0%
gso_004_input4ranker_selected['baseline', 'ranker_selected', 'repair_hybrid']0.2448-0.0006-0.000410.0032-1185334100.0%
gso_005_input4ranker_selected['baseline', 'ranker_selected', 'repair_hybrid']0.17340.00000.00119-0.00799052100.0%
gso_006_input4baseline['baseline']-0.02280.00000.000000.000000.0%
gso_007_input4ranker_selected['baseline', 'ranker_selected', 'repair_hybrid']0.31700.00130.00281-0.0188-258028100.0%
gso_008_input4baseline['baseline']-0.10430.00000.000000.000000.0%
gso_009_input4ranker_selected['baseline', 'ranker_selected', 'repair_hybrid']0.2533-0.0027-0.00006-0.0029-677274100.0%

选择来源统计

候选来源次数比例
baseline220.0%
ranker_selected880.0%

可视化结果

A18 silhouette IoU by method
A18 silhouette IoU by method
A18 Chamfer by method
A18 Chamfer by method
A18 watertight rate by method
A18 watertight rate by method
A18 predicted faces by method
A18 predicted faces by method
Chamfer Δ
-0.00029

负数表示比 baseline 略好,是 A15-A18 中第一次同时结合 view consistency 后得到的 Chamfer 正向均值。

IoU Δ
0.0159

输入视角 silhouette 一致性高于 baseline。

faces Δ
-567739

面片数显著下降,但水密率仍需提升。

实验结论

下一步想法