3D shapes from scanning, reconstruction, or AI-generated content often lack simple quad mesh layouts—critical for efficient editing and modeling. Existing quad-remeshing techniques typically produce complex layouts with irregular loops, leading to tedious manual cleanup and extensive algorithm tuning. We introduce SQUADGEN, a diffusion-based generative framework that leverages Chart Distance Fields (CDF) to synthesize simple quad layouts on 3D shapes. Our approach addresses two key challenges: (1) the discrete nature of mesh connectivity, which hinders learning, and (2) the scarcity of large-scale datasets with simple quad meshes. To overcome the first, we propose CDF, a continuous surface-based representation enabling effective learning and synthesis of quad layouts. To address the second, we define loop-aware simplicity metrics and construct a large-scale dataset of high-quality quad layouts recovered from public 3D repositories through a robust quad-recovery pipeline. Extensive evaluations across diverse 3D inputs show that SQUADGEN consistently outperforms existing methods, producing robust, artist-friendly simple quad layouts.

Synthesizing realistic 3D indoor scenes remains challenging due to data scarcity and the difficulty of simultaneously enforcing global architectural constraints and local semantic consistency. Existing approaches often overlook structural boundaries or rely on fully connected relation graphs that introduce redundant generation errors. Inspired by human design cognition, we present CasLayout, a cascaded diffusion framework that decomposes the joint scene generation task into four conditional sub-stages with explicit physical and semantic roles: (1) predicting furniture quantity and categories, (2) refining object sizes and feature embeddings, (3) modeling spatial relationships in a latent space, and (4) generating Oriented Bounding Boxes (OBBs). This decoupled architecture reduces data requirements and enables flexible integration of Large Language Models (LLMs) and Vision Language Models (VLMs) for zero-shot tasks such as image-to-scene generation. To maintain physical validity within complex floor plans, we explicitly model building elements ( e.g. , walls, doors, and windows) as conditional constraints. Furthermore, to address the high entropy of dense relation graphs, we introduce a sparse relation graph formulation aligned with human spatial descriptions. By encoding these sparse graphs into a compact latent space using a bidirectional Variational Autoencoder (VAE), the proposed framework provides enhanced relational controllability, allowing generated layouts to better respect functional organization. Experiments demonstrate that CasLayout achieves state-of-the-art performance in fidelity and diversity while enabling improved controllability in practical applications.

A promising direction for pre-training 3D point clouds is to leverage the massive amount of data in 2D, whereas the domain gap between 2D and 3D creates a fundamental challenge. This paper proposes a novel approach to point-cloud pre-training that learns 3D representations by leveraging pre-trained 2D networks. Different from the popular practice of predicting 2D features first and then obtaining 3D features through dimensionality lifting, our approach directly uses a 3D network for feature extraction. We train the 3D feature extraction network with the help of the novel 2D knowledge transfer loss, which enforces the 2D projections of the 3D feature to be consistent with the output of pre-trained 2D networks. To prevent the feature from discarding 3D signals, we introduce the multi-view consistency loss that additionally encourages the projected 2D feature representations to capture pixel-wise correspondences across different views. Such correspondences induce 3D geometry and effectively retain 3D features in the projected 2D features. Experimental results demonstrate that our pre-trained model can be successfully transferred to various downstream tasks, including 3D shape classification, part segmentation, 3D object detection, and semantic segmentation, achieving state-of-the-art performance.

Abstract Multi‐bounce, pure specular light paths produce complex lighting effects, such as caustics and sparkle highlights, which are challenging to render due to their sparse and diverse nature. We introduce a learning‐based method for the efficient rendering of pure specular light transport. The key idea is training a neural network to model the distribution of all specular light paths between pairs of endpoints for one specular object. To achieve this, for each object, our method models the distribution of sparse and diverse specular light paths between two endpoints using smooth 2D maps of ray directions from one endpoint and represents these maps with a 2D convolutional network. We design a training scheme to efficiently sample specular light paths from the scene and train the network. Once trained, our method predicts specular light paths for a given pair of endpoints using the network and employs root‐finding‐based algorithms for rendering the specular light transport. Experimental results demonstrate that our method generates high‐quality results, supports dynamic lighting and moving objects within the scene, and significantly enhances the rendering speed of existing techniques.

To aggregate useful information among diversified sources, a hotspot research topic called truth discovery has emerged in recent years. Existing truth discovery methods attempt to infer the true attribute values for the entities by identifying and trusting reliable data sources. That is, the values provided by reliable sources are more likely to be the true values. However, all these methods neglect the relations among different entities, which play important roles in truth discovery task. When reliable data sources cannot provide sufficient information of entities, the true attribute values of these entities can still be inferred by propagating trustworthy information from related entities. Motivated by this, in this paper, we introduce theconstrained truth discoveryproblem. We incorporate denial constraints, a universally quantified first-order logic formalism which can express a large number of effective and widely existing relations among entities, into the process of truth discovery. We formulate it as a constrained optimization problem and analyze its hardness. To address the problem, we propose algorithms to partition the entities into disjoint groups, and generate arithmetic constraints for each disjoint group separately. Then, the true attribute values of the entities in each disjoint group are derived by minimizing the objective function under the corresponding arithmetic constraints. Experimental results on both real-world and synthetic datasets demonstrate that the proposed approach achieves good performance even with very few constraints and reliable sources.

Aggregating the information provided by multiple data sources, which is also known as information integration , plays an important role in data analytics. Since there often exists recording errors, intentional errors, conflicts and outdated data across different data sources, finding the true attribute values of each entity is a fundamental task of crucial importance [3] . The process to fulfill this task is called truth discovery , which has been extensively studied in the literature.