arXiv subjects

Laurent Guigues

Publications and source records attributed to Laurent Guigues.

3 recordsLinked to original sources

arXiv2026

HAD: Hallucination-Aware Diffusion Priors for 3D Reconstruction

Diffusion priors have recently demonstrated strong capability in enhancing the quality of sparse-view 3D reconstruction by augmenting training views at novel viewpoints, but they inevitably introduce hallucinated content -- artifacts inconsistent with the input views -- into the final 3D model. To address this challenge, we propose Hallucination-Aware Diffusion prior (HAD), which estimates pixel-wise hallucination score maps for augmented images by leveraging multi-view reasoning capabilities from a feedforward novel view synthesis (NVS) network pre-trained on large-scale 3D data. These hallucination scores enable selective masking of unreliable pixels during the progressive 3D reconstruction procedure, preventing the introduction of non-existent artifacts into the 3D model. To further enhance performance, we create multiple versions of augmented images at each novel view by conditioning the diffusion prior on different input views, which are then fused into a final image that leverages the broader context across all input views. We show that our method substantially reduces hallucination artifacts in diffusion-assisted 3D reconstruction, thereby achieving state-of-the-art performance across multiple benchmarks on novel view synthesis. Our project are publicly available at \href{https://xiliu8006.github.io/HAD-Project-website/}{project website}.

cs.CV↗

arXiv2026

Graph Normalization: Fast Binarizing Dynamics for Differentiable MWIS

We introduce Graph Normalization (GN), a principled dynamical system on graphs that serves as a differentiable approximation engine for the NP-hard Maximum Weight Independent Set (MWIS) problem. MWIS encompasses many combinatorial challenges, including optimal assignment, scheduling, set packing, and MAP inference in discrete Markov Random Fields. Unlike Belief Propagation, we prove GN always converges to a binary indicator of a Maximum Independent Set. GN realizes a fast quasi-Newton descent through an exact Majorization-Minimization step, systematically improving the MWIS relaxed primal objective. We establish an equivalence between GN and the Replicator Dynamics of a nonlinear evolutionary game, where vertices compete for inclusion in an independent set. While a non-potential game, the GN game follows Fisher's Fundamental Theorem of Natural Selection, where the average fitness equals the MWIS primal objective and strictly increases. This connection leads to a weighted extension of the Motzkin-Straus theorem, showing MISes are in bijection with the local minima of a quadratic form over a tilted simplex. For the Assignment Problem, GN acts as a variant of the Sinkhorn algorithm that naturally converges to a hard assignment while generalizing to arbitrary constraint graphs. We demonstrate GN's performance as a fast binarization engine for the state-of-the-art Bregman-Sinkhorn relaxed MWIS solver. On real-world benchmarks with up to 1M edges, GN identifies solutions within 1% of the best known results in seconds on a CPU. GN opens new avenues for deep learning architectures requiring differentiable, "hard" decisions under constraints, with applications in structured sparse attention, dynamic network pruning, and Mixture-of-Experts. Beyond core AI, the GN framework enables end-to-end learning of constrained optimization in computer vision, computational biology, and resource allocation.

cs.LG↗

arXiv2020

Concerning Iterative Graph Normalization and Maximum Weight Independent Sets

We consider a very simple dynamical system on weighted graphs which we call Iterative Graph Normalization (IGN) and a variant in which we apply a non-linear activation function to the weights after each normalization. We show that the indicator vectors of the Maximal Independent Sets of the graph are the only binary fixed points of IGN, that they are attractive under simple conditions on the activation function and we characterize their basins of attraction. We enumerate a number of other fixed points and we prove repulsivity for some classes. Based on extensive experiments and different theoretical arguments we conjecture that IGN always converges and converges to a binary solution for non-linear activations. If our conjectures are correct, IGN would thus be a differentiable approximation algorithm for the Maximum Weight Independent Set problem (MWIS), a central NP-hard optimization problem with numerous applications. IGN is closely related to a greedy approximation algorithm of MWIS by Kako et al. which has a proven approximation ratio. Experimental results show that IGN provides solutions of very similar quality. In the context of the Assignment Problem, IGN corresponds to an iterative matrix normalization scheme which is closely related to the Sinkhorn-Knopp algorithm except that it projects to a permutation matrix instead of a doubly stochastic matrix. We relate our scheme to the Softassign algorithm and provide comparative results. As Graph Normalization is differentiable, its iterations can be embedded into a machine learning framework and used to train end-to-end any model which includes a graphical optimization step which can be cast as a maximum weight independent set problem. This includes problems such as graph and hypergraph matching, sequence alignment, clustering, ranking, etc. with applications in multiple domains.

cs.DM↗