24 March 2025 in IEEE Transactions on Artificial Intelligence
Point clouds captured from 3D scanners are often sparse and incomplete due to occlusions, limited viewpoints, and sensor constraints. These limitations hinder applications in robotics, autonomous navigation, and augmented reality. Hence, point cloud completion is crucial for generating reliable 3D object representations. Existing methods often struggle to capture structural patterns effectively, which leads to low-quality reconstructions. To address these challenges, we propose Centroid Transformer (CenFormer), a novel transformer-based network for point cloud completion. CenFormer introduces two distinct types of centroids, namely Preserved and Dispersed, to facilitate fine-grained reconstruction. The proposed design includes three innovations: i) a Centroid Generation Block to aggregate features for preserved centroids, ii) a Centroid Dispersion Block to predict offsets for dispersed centroids, and iii) a Fine-grained Point Generation Block to refine local patterns around centroids. These components jointly enable the network to effectively capture local structural details and strategically target missing regions for fine-grained 3D shape reconstruction. Experiments on various benchmark datasets demonstrate that CenFormer significantly outperforms state-of-the-art methods in both visualization results and quantitative metrics.
11 September 2025 in IEEE Transactions on Artificial Intelligence
Effective camouflage is needed for defense personnel and assets to blend seamlessly with the complex and dynamic environments. The camouflage technique must maintain the concealment capability across various environments and geographic regions. However, existing approaches, including manual design, computer-aided techniques, and deep learning methods, face significant challenges in achieving automation, scalability, and generalization across diverse and uncalibrated scenes. To address these limitations, we propose a novel diffusion-based method with few-shot learning to generate environment-guided camouflage patterns. Our method, called CamoX, consists of two major stages: meta learning and few-shot learning. In the meta learning stage, our method introduces a latent diffusion-based architecture that automatically generates camouflage patterns from noise, eliminating manual intervention and enabling scalable production. In the few-shot learning stage, our approach enforces similarity between the latent features of the camouflage patterns and target scenes by optimizing the guided mean absolute error loss. This innovation allows the generated camouflage patterns to adapt seamlessly to multiple environments with minimal retraining. Furthermore, this paper introduces a comprehensive camouflage dataset, called Camo-Meta, comprising 144,750 realistic camouflage patterns and associated metadata to support research in camouflage generation. Experimental results on multiple datasets demonstrate that CamoX outperforms existing state-of-the-art methods in key metrics.
1 January 2026 in IEEE Transactions on Instrumentation and Measurement
31 January 2025 in TechRxiv
Camouflage is an adaptive strategy that allows organisms and objects to blend with their surrounding environments, thereby evading detection by predators or adversaries. Current artificial methods, including manual painting, computer-aided techniques, and deep learning approaches, face significant challenges related to manual intervention, scalability, and generalization across diverse and non-calibrated scenes. In this paper, we propose a novel few-shot diffusion-based method for camouflage pattern generation, called Camouflage Diffusion (CamoDiff). Our method consists of two distinct stages: meta learning and few-shot learning. During the meta learning stage, our method employs a diffusion-based architecture to automatically generate camouflage patterns from noise, thus it eliminates manual intervention and enables scalable production. Additionally, our approach integrates the guidance mean absolute error (GMAE) loss in the few-shot learning stage. This allows the generated camouflage to adapt seamlessly to new environments with minimal retraining, regardless of varying viewpoints. We also introduce a comprehensive camouflage dataset and generation tool, which can be used as benchmarks for future research. Experimental results demonstrate that CamoDiff outperforms existing state-of-the-art methods in camouflage pattern generation on different datasets and metrics.