Learning semantic representations from point sets of 3D object shapes is often challenged by significant geometric variations, primarily due to differences in data acquisition methods. Typically, training data is generated using point simulators, while testing data is collected with distinct 3D sensors, leading to a simulation-to-reality (Sim2Real) domain gap that limits the generalization ability of point classifiers. Current unsupervised domain adaptation (UDA) techniques struggle with this gap, as they often lack robust, domain-insensitive descriptors capable of capturing global topological information, resulting in overfitting to the limited semantic patterns of the source domain. To address this issue, we introduce a novel Topology-Aware Modeling (TAM) framework for Sim2Real UDA on object point clouds. Our approach mitigates the domain gap by leveraging global spatial topology, characterized by low-level, high-frequency 3D structures, and by modeling the topological relations of local geometric features through a novel self-supervised learning task. Additionally, we propose an advanced self-training strategy that combines cross-domain contrastive learning with self-training, effectively reducing the impact of noisy pseudo-labels and enhancing the robustness of the adaptation process. Experimental results on three public Sim2Real benchmarks validate the effectiveness of our TAM framework, showing consistent improvements over state-of-the-art methods across all evaluated tasks. The source code of this work will be available athttps://github.com/zou-longkun/TAG.git.

We introduce ArcPro, a novel learning framework built on architectural programs to recover structured 3D abstractions from highly sparse and low-quality point clouds. Specifically, we design a domain-specific language (DSL) to hierarchically represent building structures as a program, which can be efficiently converted into a mesh. We bridge feedforward and inverse procedural modeling by using a feedforward process for training data synthesis, allowing the network to make reverse predictions. We train an encoder-decoder on the points-program pairs to establish a mapping from unstructured point clouds to architectural programs, where a 3D convolutional encoder extracts point cloud features and a transformer decoder autoregressively predicts the programs in a tokenized form. Inference by our method is highly efficient and produces plausible and faithful 3D abstractions. Comprehensive experiments demonstrate that ArcPro outperforms both traditional architectural proxy reconstruction and learning-based abstraction methods. We further explore its potential to work with multi-view image and natural language inputs. Project page: https://vcc.tech/research/2025/ArcPro.

The Low-Altitude Intelligent Network (LAIN) serves as a critical enabler for low-altitude economic development, showing significant potential in areas such as environmental monitoring. These applications typically require extensive computational resources to drive large-scale models, posing challenges to resource-constrained aerial platforms like unmanned aerial vehicles (UAVs). Although air-ground model collaboration has been studied by some early studies, most approaches assume homogeneous computational resources, overlooking the heterogeneity among UAVs as end nodes. To address this challenge, we introduce an air-ground model collaboration framework that considers heterogeneous computational resources among multiple UAVs and limited wireless transmission bandwidth between UAVs and ground servers. In this framework, UAVs working as end nodes handle data collection and local inference with small models, while the edge servers on the ground perform large model inference and updates. We propose a joint optimization strategy that optimizes both data transmission and resource allocation, with the goal of improving the framework’s inference accuracy by maximizing the mean average precision (mAP). Simulations on object detection tasks show that our framework outperforms existing methods under different communication bandwidths and data scales.

A radio map captures the spatial distribution of wireless channel parameters, such as the strength of the signal received, across a geographic area. The problem of fine-grained three-dimensional (3D) radio map estimation aims to infer a high-resolution radio map for the two-dimensional (2D) area at an arbitrary target height within a 3D region of interest, using radio samples collected by sensors sparsely distributed in that 3D region. Solutions to the problem are crucial for the improvement in drone spectrum utilization in the rapidly developing low-altitude economy. However, this problem is challenging due to ultra-sparse sampling, where the number of collected radio samples is far fewer than the desired resolution of the radio map to be estimated. In this paper, we design a generative framework called GenRadio which employs the creative power of generative AI to address the ultra-sparse sampling challenge and solve the problem. GenRadio consists of three key blocks: (1) an augmentation block that uses radio propagation models to project samples collected at different heights to the 2D area at the target height; (2) a generation block that employs a diffusion model under a Mixture of Experts (MoE) architecture to generate a diverse set of fine-grained radio map candidates; and (3) an election block that leverages radio propagation models to identify the best map candidate from the diverse set. Experimental results demonstrate that GenRadio efficiently solves the fine-grained 3D radio map estimation problem from an ultra-low sampling rate of 0.1%, and significantly outperforms state-of-the-art.

In urban environments, numerous IoT devices rely on accurate radio map information to enable critical applications such as localization, trajectory planning, and communication. The effectiveness of these applications hinges on the availability of high-fidelity and spatially comprehensive radio maps. To address this, we propose an optimal path planning approach with distance constraints for efficient sampling in urban radio map reconstruction. Unlike traditional random sampling strategies, our method restricts sampling points to lie along a feasible road network and imposes a fixed distance constraint on the sampling path. We design a structure-aware genetic algorithm (GA_s) to optimize the path for mobile sampling, incorporating an unsupervised population evaluation metric to assess the fitness of candidate solutions. Experimental results show that, under equal sampling distance conditions, GA_s achieves a Root Mean Square Error (RMSE) of 0.0626 in radio map recovery—outperforming random sampling (0.0796), A* algorithm (0.0748), and a conventional genetic algorithm (GA_c). These results demonstrate the effectiveness of our method in enabling efficient and accurate radio map reconstruction for mobile sampling platforms in real-world urban settings.

Global Navigation Satellite Systems (GNSS) and base station (BS)-based wireless positioning are the predominant technologies for outdoor user equipment (UE) localization. However, their performance deteriorates significantly in dense urban environments with complex architectural structures. This degradation arises because obstructions disrupt line-of-sight (LoS) conditions between the UE and satellites or base stations, reducing positioning accuracy. To address this challenge, this paper introduces LocVMunet, a novel deep neural network designed for high-precision localization based on received signal strength (RSS) from a limited number of base stations. The proposed LocVMunet seamlessly integrates the Cross-Scan Module (CSM) with signal distribution modeling, enhancing spatial feature extraction and improving localization accuracy. Unlike traditional methods, this neural network-based approach inherently mitigates the impact of non-line-of-sight (NLOS) conditions, enabling real-time localization across any region covered by the radio map. Experimental evaluations demonstrate that the LocVMunet establishes a new benchmark for RSS-based urban localization, significantly outperforming state-of-the-art neural network-based methods. Notably, it achieves a 24.8% positioning error reduction over LocUNet and a 22.7% improvement over LocSwinUnet in the five-base-station setup, highlighting its robustness and effectiveness in urban positioning.

Radio map presents communication parameters of interest, e.g., radio power, across a geographical region in a specific frequency band. It can be leveraged to improve the efficiency of spectrum utilization. The problem of constructing a radio map involves utilizing measurements from sparsely distributed sensors to infer the parameter of interest at every point in the region. One major challenge of solving this problem is ensuring a strong generalization capability for the solution, i.e., ensuring that the solution is able to make accurate inferences even in the presence of unknown system variables that can affect radio propagation, such as obstacle layouts and weather conditions. This paper focuses on analyzing and optimizing the generalization capability of radio map construction, an aspect that has been neglected in prior research as far as we know. We design RobUNet—a UNet-based algorithm that captures multi-scale features of radio maps. It further leverages mechanisms of residual connection, channel attention, and pixel attention to enhance its generalization capability. Simulations based on real-world dataset show that (i) the radio maps constructed by RobUNet are much more accurate than those constructed by many existing solutions in the presence of unknown system variables, and (ii) the accuracy of the radio maps constructed by RobUNet when system variables are unknown is close to that of the radio maps constructed by RobUNet when system variables are known. As a result, the generalization capability of RobUNet is much stronger than that of state-of-the-art.

Model compression and distillation techniques have become essential for deploying deep learning models efficiently. However, existing methods often encounter challenges related to model generalization and scalability for harnessing the expertise of pre-trained large models. This paper introduces CoTuning, a novel framework designed to enhance the generalization ability of neural networks by leveraging collaborative learning between large and small models. CoTuning overcomes the limitations of traditional compression and distillation techniques by introducing strategies for knowledge exchange and simultaneous optimization. Our framework comprises an adapter-based co-tuning mechanism between cloud and edge models, a scale-shift projection for feature alignment, and a novel collaborative knowledge distillation mechanism for domain-agnostic tasks. Extensive experiments conducted on various benchmark datasets demonstrate the effectiveness of CoTuning in improving model generalization while maintaining computational efficiency and scalability. The proposed framework exhibits a significant advancement in model compression and distillation, with broad implications for research in the collaborative evolution of large-small models.

Fine-grained visual classification can be addressed by deep representation learning under supervision of manually pre-defined targets (e.g., one-hot or the Hadamard codes). Such target coding schemes are less flexible to model inter-class correlation and are sensitive to sparse and imbalanced data distribution as well. In light of this, this paper introduces a novel target coding scheme - dynamic target relation graphs (DTRG), which, as an auxiliary feature regularization, is a self-generated structural output to be mapped from input images. Specifically, online computation of class-level feature centers is designed to generate cross-category distance in the representation space, which can thus be depicted by a dynamic graph in a non-parametric manner. Explicitly minimizing intra-class feature variations anchored on those class-level centers can encourage learning of discriminative features. Moreover, owing to exploiting inter-class dependency, the proposed target graphs can alleviate data sparsity and imbalanceness in representation learning. Inspired by recent success of the mixup style data augmentation, this paper introduces randomness into soft construction of dynamic target relation graphs to further explore relation diversity of target classes. Experimental results can demonstrate the effectiveness of our method on a number of diverse benchmarks of multiple visual classification, especially achieving the state-of-the-art performance on three popular fine-grained object benchmarks and superior robustness against sparse and imbalanced data. Source codes are made publicly available at https://github.com/AkonLau/DTRG.