AI 中文总结
研究摄影测量重建在无地面真值时估计自身可靠性的问题，提出无轨迹泄漏留出协议，通过特定图像子集和3D点重新定位计算mAA置信度，发现该协议能测内部几何一致性，但非绝对精度，不可替代控制点评估和严重失败门限。
AI 中文摘要
自动摄影测量检查从3D重建中发出度量测量值，而在没有外部测量的情况下，其自身的正确性通常是未知的。重建能否在没有地面真值的情况下估计自身的可靠性？我们形式化了一种无轨迹泄漏留出协议：保留图像的一个确定性子集，并仅针对至少两个保留图像看到的3D点进行重新定位——这是一个轨迹级障碍，因此一个视图永远不会针对它帮助创建的结构进行测试——汇总为mAA置信度。留出的观测值仍进入构建可信结构的光束平差，因此我们称其为无轨迹泄漏（一种更严格的重新映射变体在3/5度时证实了这一点）。在GNSS参考捕获以及ETH3D、EuRoC和IMC 2025基准测试中，我们报告了四个发现。（i）该协议在计算上是合理的：良好的重建得分可达毫度级的自一致性。（ii）阈值化的自一致性达到饱和，不跟踪绝对精度：置信度保持在1.00附近，而真实RTK误差在一次捕获中波动高达14倍，并且每次捕获的相关性在五次捕获中符号不稳定（-0.57至+0.98；合并的95%置信区间跨越零），因此它无法控制精度。（iii）它仅在失败破坏内部一致性时标记严重失败：一个碎片化模型会降低置信度，但一个单一的自一致、全局扭曲的模型会避开它——四次捕获中有三次给出了一个置信度为1.00但错误达55 - 106米的模型。（iv）在独立的ETH3D和IMC 2025地面真值上也存在同样的二分法。无轨迹泄漏留出法测量的是内部几何一致性，而非绝对精度：既不能替代控制点评估，也不是一般的严重失败门限。我们发布了该协议和降级工具。
英文摘要
Internal self-consistency cannot certify the accuracy of a photogrammetric reconstruction, and the failure is structural rather than a matter of tuning. This matters because hold-out self-validation scores are increasingly offered as quality evidence for metric deliverables whose correctness is otherwise unknown without an external survey. We formalise a track-leakage-free hold-out protocol: a deterministic image subset is withheld, and each withheld view is re-localised against only those 3D points supported by two or more retained images, so no view is tested against structure it helped create. We evaluate it on five GNSS-referenced captures across four sites, 13 ETH3D scenes, a EuRoC flight and 30 IMC 2025 scenes. The protocol is well-posed but does not measure accuracy. It saturates: the internal confidence score stays pinned at 1.00 while true error swings 14.1x within one capture. It is blind to coherent distortion: fragmenting corruption is caught, but internally self-consistent, globally distorted models are not, and were wrong by 55-106 m at confidence 1.00 at three of four captures. On IMC 2025 it separates failed from successful reconstructions (rho = 0.68) yet ranks nothing among the successful (rho = 0.01). Track-leakage-free hold-out measures internal geometric consistency: a fragmentation warning, not a substitute for control-point accuracy assessment.
Comments16 pages, 4 figures. v4: retitled to lead with the finding, and the abstract rewritten accordingly. No change to the results, methods, data or conclusions. Code, harness and result tables archived at Zenodo (doi:10.5281/zenodo.21737748)
