实验 30:A24 Calibrated VGGT Geometry Gate

把 VGGT depth 作为硬锚点、normal 作为软证据,校准无 GT repair 路由

实验定位 混合偏正 / 校准有效但仍保守

A24 不是重新生成 mesh,而是在 A23 已得到的 VGGT depth/normal 指标上做 retrospective router calibration:选择阶段只使用 VGGT 和 A20 visible-safe 信号,GT Chamfer/F-score 只在最后用来判断路由有没有误拒或误接收。

关键变化是把 A23 的 normal hard reject 降级为 soft evidence。结果:A23 hard gate 只选择 2 个 repair,false reject=5;calibrated gate 选择 3 个 repair,false reject 降到 4,并救回了 gso_002,同时仍拒绝 gso_008 这类明显坏 repair。

实验设计(Image-2 风格)

实验设计图:A24 calibrated VGGT geometry gate
实验设计图:A24 calibrated VGGT geometry gate
模块设计图:depth-first / normal-soft router
模块设计图:depth-first / normal-soft router

模块设计(Image-2 风格)

Hypothesis

Depth:VGGT depth is treated as the hard visible geometry anchor after A23 showed it catches gso_006/gso_008.

Normal:VGGT normals from finite-difference world points are noisy; they should be soft evidence and only reject when the normal drop is very large without depth support.

Visible-safe:A20 visible-safe remains useful, but it should not veto a repair when depth strongly supports the repair.

Compared Routers

  • a23_hard_gate:A23 hard: depth/normal/A20 任一拒绝即回退
  • depth_only:只用 VGGT depth hard reject,normal 不否决
  • depth_visible_safe:VGGT depth + A20 visible-safe 补充约束
  • calibrated_depth_normal:depth-first + normal-soft calibrated gate

实验结果(表格)

策略选择 repair 数false rejectfalse acceptselected Chamfer Δselected F@5 Δ选择的 repair case
A23 hard gate250-0.000170.0017gso_000_input4, gso_009_input4
Depth only812-0.000290.0048gso_000_input4, gso_001_input4, gso_002_input4, gso_003_input4, gso_004_input4, gso_005_input4, gso_007_input4, gso_009_input4
Depth + visible-safe340-0.000540.0067gso_000_input4, gso_002_input4, gso_009_input4
Calibrated depth-normal340-0.000540.0067gso_000_input4, gso_002_input4, gso_009_input4

Calibrated gate 逐 case 决策

caseselectedscoreVGGT depth ΔVGGT normal ΔGT Chamfer ΔGT F@5 Δ回顾标签原因
gso_000_input4repair_hybrid0.02323-0.02330-0.0024-0.001600.0202ok[]
gso_001_input4baseline0.00011-0.00020-0.0031-0.001050.0082false reject['visible-safe rejects and depth does not compensate']
gso_002_input4repair_hybrid0.00440-0.00656-0.0719-0.003770.0502ok[]
gso_003_input4baseline-0.002010.00193-0.0025-0.00005-0.0041false reject['visible-safe rejects and depth does not compensate']
gso_004_input4baseline0.00004-0.00009-0.0019-0.000410.0032false reject['visible-safe rejects and depth does not compensate']
gso_005_input4baseline0.000040.000080.00110.00119-0.0079ok['visible-safe rejects and depth does not compensate']
gso_006_input4baseline-0.024150.031910.07750.001440.0204false reject['depth worsens beyond hard threshold', 'visible-safe rejects and depth does not compensate']
gso_007_input4baseline0.000820.001940.02770.00281-0.0188ok['visible-safe rejects and depth does not compensate']
gso_008_input4baseline-0.013450.015190.01740.00474-0.0168ok['depth worsens beyond hard threshold', 'visible-safe rejects and depth does not compensate']
gso_009_input4repair_hybrid0.000350.000170.0052-0.00006-0.0029ok[]

可视化结果

四种路由策略的 false reject / false accept 对比
四种路由策略的 false reject / false accept 对比
四种路由策略的回顾质量对比
四种路由策略的回顾质量对比

实验结论

下一步想法