--- title: Conditional Geometric Reconstruction url: https://www.emergentmind.com/topics/conditional-geometric-reconstruction type: topic --- # Conditional Geometric Reconstruction Searching arXiv for recent papers relevant to conditional geometric reconstruction. Tool call: arxiv_search with query "conditional geometric reconstruction 3D reconstruction diffusion sparse views canonical anchors geometry" Conditional geometric reconstruction can be understood as a family of reconstruction problems in which the target geometry is inferred under explicit conditioning by auxiliary observations, latent variables, acquisition geometry, temporal context, or learned priors. In the literature, this conditioning ranges from classical formulations with uncertain projection geometry, sparse cross-sections, and point–normal measurements to contemporary feed-forward and diffusion-based systems that condition on sparse views, canonical anchors, time, age, ray bundles, or generative intermediates. Across these settings, the central technical objective is similar: to reduce ambiguity in ill-posed inverse problems by coupling geometric inference to additional structure rather than reconstructing shape from raw observations alone [2203.01045], [2412.08412], [2503.22349]. ## 1. Conceptual scope and historical lineages Early work framed reconstruction conditionally in explicitly geometric terms. In “Geometric reconstruction from point-normal data” [1003.3499], the input is a sparse set of point–normal samples \[ \{ (p_i, n_i) \in \mathbb{R}^3 \times S^2 : i=1 \ldots N \}, \] and the goal is to recover planar faces and combinatorial adjacencies consistent with those samples. For a convex polyhedron with one marker per face, the reconstruction is given by the intersection of half-spaces \[ H_i = \{ x \in \mathbb{R}^3 : (x - p_i)\cdot n_i \le 0 \}, \] which yields exactly the polyhedron [1003.3499]. In “Geometric Tomography With Topological Guarantees” [1007.2452], reconstruction is conditioned on a family of cutting planes and their cross-sections, and the reconstructed object is defined by a nearest-point rule relative to the arrangement cells of those planes. Under the Density Condition \[ h_C < \mathrm{reach}_C(O) \] and the Transversality Condition \[ h_C < \tfrac12(1-\sin \alpha_C)\,\mathrm{reach}_C(O), \] the output preserves the homotopy type, and is further shown to be homeomorphic and isotopic to the original object [1007.2452]. A second lineage arises in inverse problems with uncertain acquisition geometry. In “A Bayesian Approach to CT Reconstruction with Uncertain Geometry” [2203.01045], the unknown image \(x\) and the unknown geometry parameters \(\theta\) are reconstructed jointly under the forward model \[ y = A(\theta)x + \epsilon, \qquad \epsilon \sim N(0,\lambda^{-1}I). \] Here the condition is not merely the measured sinogram \(y\), but the joint probabilistic structure tying image, projection geometry, and hyperparameters into one posterior [2203.01045]. Recent deep-learning work generalizes this idea from explicit geometry to learned conditioning. “Conditional Single-view Shape Generation for Multi-view Stereo Reconstruction” [1904.06699] models \(p(S\mid I)\) with a latent variable \(r\sim \mathcal N(0,I_d)\), while “Pragmatist” [2412.08412] reformulates reconstruction from sparse, unposed observations as conditional novel view synthesis, using generated complete observations to facilitate subsequent reconstruction. This suggests that contemporary conditional reconstruction inherits two older themes: uncertainty modeling and geometric regularization, but relocates them into learned representations and generative priors. ## 2. Mathematical forms of conditioning The literature instantiates conditioning in several mathematically distinct ways. A probabilistic formulation appears in Bayesian CT, where the posterior \[ p(x,\theta,\lambda,\delta \mid y) \propto p(y\mid x,\theta,\lambda)\,p(x\mid \delta)\,p(\theta)\,p(\lambda)\,p(\delta) \] couples reconstruction and geometry estimation directly [2203.01045]. The key conditional density, \[ p(y\mid x,\theta,\lambda) \propto \lambda^{m/2}\exp\!\left[-(\lambda/2)\|A(\theta)x-y\|_2^2\right], \] makes the reconstruction explicitly dependent on uncertain geometry [2203.01045]. A generative conditional formulation appears in single-view shape generation. There the target is the conditional density \[ p(S\mid I), \] with a latent-conditioned generator \[ S=f_\theta(I,r), \qquad r\sim \mathcal N(0,I_d), \] so that multi-view reconstruction is posed as the intersection of the per-view shape manifolds \[ \Omega(I_i)=\{f_\theta(I_i,r)\mid r\in \mathbb{R}^{128}\}, \qquad \bigcap_{i=1}^n \Omega(I_i) \] [1904.06699]. In this setting, conditioning is used to represent uncertainty in unseen geometry rather than to collapse it into a single deterministic predictor. Conditioning can also be temporal or demographic. In “Conditional Temporal Attention Network (CoTAN)” [2307.11870], the time-varying velocity field is conditioned jointly on integration time \(t\) and post-menstrual age \(a\). The attention weights \[ \alpha(t,a)=\mathrm{softmax}(f(t,a)) \] define a conditional time-varying velocity field \[ v_t(x;a)=\sum_{r=1}^R \sum_{m=1}^M \alpha^{r,m}(t,a)\,v^{r,m}(x), \] which is then integrated to deform a template mesh diffeomorphically [2307.11870]. In dynamic monocular 4D reconstruction, conditioning appears as arbitrary spatio-temporal querying. “4RC” [2602.10094] encodes the entire video once, \[ z=E(V), \] and answers geometry–motion queries with \[ (G(x), M(x,\tau))=D(z;x,\tau), \] using the factorization \[ P_i^{t_i\to \tau}(x)=P_i^{t_i}(x)+\Delta P_i^{t_i\to \tau}(x) \] to separate base geometry from time-dependent relative motion [2602.10094]. A plausible implication is that conditional geometric reconstruction is not tied to a single output geometry; it can also denote conditional access to geometry across time, view, or deformation state. ## 3. Representations and conditional carriers Conditioning is realized through the representation chosen for geometry. Several recent systems use explicit geometric carriers whose parameters are then modulated by learned context. In “360-GeoGS” [2601.02102], a single 360° RGB panorama is processed by a SphereCNN backbone to produce a spherical cost volume, while high-level image features are aggregated into a global conditioning code \(C_{\mathrm{cond}}\). Feature-wise Linear Modulation applies \[ \hat F=\gamma(C_{\mathrm{cond}})\cdot F+\beta(C_{\mathrm{cond}}) \] before a U-Net decoder regresses per-pixel 3D Gaussian parameters at panorama resolution \(512\times 1024\) [2601.02102]. The Gaussian covariance is parameterized as \[ \Sigma = R(\theta)^\top \,\mathrm{diag}(s)\,R(\theta), \] which makes position, scale, and rotation available for direct geometric regularization [2601.02102]. A different explicit carrier is the triplane signed distance field. In “GCRayDiffusion” [2503.22349], a transformer-based image encoder produces three feature planes \(\{F_x,F_y,F_z\}\), and the SDF is decoded as \[ F_\theta(x)=\mathcal D(f(x)). \] Camera poses are over-parameterized as neural bundle rays \[ r_k^i=(v_k^i,m_k^i,d_k^i)\in \mathbb{R}^7, \] and the denoiser is conditioned on local image features and the current SDF values at ray-hit points, \[ g_\phi\Bigl(\mathcal R_t,t \,\bigm|\, F_I, F_\theta(\{r_k^d\})\Bigr)\approx \epsilon \] [2503.22349]. Here the representation itself carries the condition: the evolving SDF regularizes pose diffusion, and the denoised rays provide explicit on-surface samples back to the SDF. Conditional reconstruction can also target topological rather than purely metric structure. “ComplexGen” [2205.14573] treats a CAD boundary representation as a chain complex \[ C_2 \xrightarrow{\partial_2} C_1 \xrightarrow{\partial_1} C_0, \] with a probabilistic model \[ P(\mathcal C\mid \mathcal P) \] for vertices, edges, faces, incidences, and geometric parameters conditioned on an input point cloud \(\mathcal P\). A tri-path transformer decoder predicts existence and incidence probabilities, and a global optimization enforces structural validness constraints such as \(\partial_1\circ \partial_2=0\) [2205.14573]. This broadens the notion of geometric reconstruction: the reconstructed quantity may be a valid B-Rep complex rather than a point cloud, mesh, or radiance field. ## 4. Coupling generative priors to explicit geometry A major contemporary direction couples conditional generation to explicit geometric reconstruction. “Pragmatist” [2412.08412] states the problem as follows: sparse, unposed observations are first completed by a multiview conditional diffusion model, then a feed-forward large reconstruction model predicts a mesh, and the poses of the input views are subsequently recovered by inverting the obtained 3D representations and using them to optimize the reconstructed object. The paper’s stated motivation is that direct prediction from unposed sparse views does not utilize geometric priors and cannot hallucinate the appearance of unseen regions, making fine geometric and textural details difficult to reconstruct [2412.08412]. “GeoRect4D” [2604.20784] addresses dynamic sparse-view 3D reconstruction by coupling an explicit dynamic 3DGS substrate with a single-step diffusion rectifier in a closed-loop optimization. The rendered frame \[ \hat I_{\mathrm{render}(t,\xi)}=\mathcal R(G(t),\xi) \] is rectified by \[ \hat I_{\mathrm{rect}(t,\xi)}=\mathcal D(\hat I_{\mathrm{render}(t,\xi)}), \] and the substrate is optimized with a hybrid objective \[ \min_G \sum_{t,v}\Big[(1-\lambda(t))\,\mathcal L_{\mathrm{data}}+\lambda(t)\,\mathcal L_{\mathrm{prior}}\Big]. \] Structural Locking injects encoder features into decoder UpBlocks, \[ h_{\mathrm{dec}}^{(l)} \leftarrow \mathrm{UpBlock}\Bigl(h_{\mathrm{dec}}^{(l-1)} + C_{\mathrm{proj}}^{(l)}(\gamma\,h_{\mathrm{enc}}^{(L-l)})\Bigr), \] while Spatio-Temporal Coordinated Attention refines latent states jointly across spatial views and short temporal windows [2604.20784]. “UniRecGen” [2604.01479] resolves the tension between deterministic reconstruction and stochastic generation by forcing both modules into a shared canonical space. A VGGT-based reconstruction mapping outputs multi-view point maps \(\{\mathbf P_i^{\mathrm{can}}\}\) in an object-centric frame, and a diffusion generator is conditioned on the same canonical anchor \(\mathcal G_{\mathrm{canon}}\). During inference, the denoiser applies \[ x_{t-1}=D_\theta(x_t,\mathcal G_{\mathrm{canon}},t), \] while the conditioning tokens are augmented as \[ \mathbf F_i^{MV}=\mathbf F_i^D+\mathrm{MLP}_{\mathrm{view}}(F_i^V)+\mathrm{MLP}_{\mathrm{cam}}(t_i^{\mathrm{cam}}) \] [2604.01479]. This suggests that conditional reconstruction increasingly depends on explicit coordinate alignment between geometric and generative modules, not merely on feature fusion. A more extreme case is “Mind the Gap: Geometrically Accurate Generative Reconstruction from Disjoint Views” [2605.07550], which defines reconstruction from disjoint inputs \(I_1,I_2\) with \(\Omega(I_1,I_2)=\varnothing\). GLADOS bridges the inputs with a generated intermediate view, globally aligns dense point maps into a coarse scaffold, and then iterates rendering, inpainting, depth estimation, and consistency optimization. Its multiview consistency term is \[ L_{\mathrm{cons}}=\sum_{i