Cloud2to3: Three-dimensional reconstruction via spherization

Cloud2to3: Three-dimensional reconstruction via spherization
Xunchuan Liu, Pak-Shing Li
arXiv:2503.19259v2 Announce Type: replace
Abstract: Inferring three-dimensional structures from two-dimensional maps remains a major challenge due to line of sight degeneracies. We present Cloud2to3, a flexible framework to reconstruct three-dimensional (3D) volumetric density fields from two-dimensional (2D) maps by generalizing the inverse Abel transform and the AVIATOR algorithm. The pipeline decomposes a 2D map into overlapping circles whose centers serve as the structural skeleton of each intensity slice. By linking these centers into a unified network, the 3D reconstruction problem is elegantly reduced to growing this skeletal tree along the line of sight. This work implements three benchmark strategies: a randomized approach, a deterministic approach, and a physically constrained model. While exact real space structural matching is degenerate due to a lack of local line of sight constraints, benchmarks show that both 2D and 3D density statistical properties are robustly preserved. The projected column density probability density function (PDF) remains invariant under multi-angle views, and testing against magnetohydrodynamic numerical simulation data confirms that texttt{Cloud2to3} can recover the intrinsic volume density statistics. Overall, our framework preserves the continuity of primary morphological patterns, allowing features like filamentary junctions and dense core clustering to be qualitatively represented within the reconstructed volume.arXiv:2503.19259v2 Announce Type: replace
Abstract: Inferring three-dimensional structures from two-dimensional maps remains a major challenge due to line of sight degeneracies. We present Cloud2to3, a flexible framework to reconstruct three-dimensional (3D) volumetric density fields from two-dimensional (2D) maps by generalizing the inverse Abel transform and the AVIATOR algorithm. The pipeline decomposes a 2D map into overlapping circles whose centers serve as the structural skeleton of each intensity slice. By linking these centers into a unified network, the 3D reconstruction problem is elegantly reduced to growing this skeletal tree along the line of sight. This work implements three benchmark strategies: a randomized approach, a deterministic approach, and a physically constrained model. While exact real space structural matching is degenerate due to a lack of local line of sight constraints, benchmarks show that both 2D and 3D density statistical properties are robustly preserved. The projected column density probability density function (PDF) remains invariant under multi-angle views, and testing against magnetohydrodynamic numerical simulation data confirms that texttt{Cloud2to3} can recover the intrinsic volume density statistics. Overall, our framework preserves the continuity of primary morphological patterns, allowing features like filamentary junctions and dense core clustering to be qualitatively represented within the reconstructed volume.