Recent advancements in 3D Gaussian Splatting (3DGS) have demonstrated remarkable rendering quality, However, their substantial computational demands hinder practical deployment on resource-constrained devices. We propose a novel plug-and-play structured compression framework that significantly reduces computational overhead while maintaining rendering fidelity. We first discover that the statistical distribution of anchor vectors is decoupled from rendering quality. Based on this finding, we propose a distribution regularization method that enforces alignment to standard Gaussian distribution through KL divergence while optimizing Gaussian radius, significantly improving entropy coding efficiency. Second, we innovatively introduce an opacity-based probabilistic pruning mechanism that transforms pruning into an opacity optimization problem, achieving intelligent scene sparsification while allowing flexible adjustment according to hardware resources. Finally, we design a lightweight high-frequency compensation network that regards the high-frequency loss caused by over-compression as a residual and effectively recovers the high-frequency details lost during the compression process through residual learning. All modules are plug-and-play and can be seamlessly integrated into mainstream structured 3DGS frameworks. Extensive experiments on Synthetic-NeRF, Tanks&Temples, Mip-NeRF360 and DeepBlending datasets demonstrate that our method significantly reduces size by over 80x compared to vanilla 3DGS while simultaneously improving fidelity. Furthermore, it achieves a better size reduction and a 20% improvement in entropy encoding efficiency when compared to HAC, while meeting the requirements for real-time rendering.

Person re-identification (reID) has received extensive study and achieved great progress in recent years. Extensive research has proved that combining global and local features is an effective solution to improve the performance of person reidentification tasks. While many existing reID approaches are still suffering from occlusions, body part missing, different lighting, and background clutter, where the learned features may achieve a sub-optimal solution. In this paper, we propose an efficient network structure, Multi-Scale Relation Network (MSRN), which can not only extract robust regional features and global features but also integrate the asymptotic cues and relations between them. In addition, we introduce a dynamic loss weight as supplementary components to improve learning efficiency and the representation capacity of our model. Extensive experiments are conducted on three widely used datasets, including Market-I501, DukeMTMC-ReID and CUHK03-NP. The experimental results indicate that our proposed method achieves the state-of-the-art results on three datasets.

This study presents a novel method for automatic detection and removal of high‐density impulse noises. The method consists of two parts: the impulse detection part and the impulse noise removal part. In impulse detection part, an automatic detector based on local mean and variance (LMVD) is presented, which can automatically pick out noisy image from massive images and output corrupted grey levels. The detector utilises LMVD of the neighbourhood of corrupted pixels to simulate the cognitive processes of human observing noisy image. In impulse noise removal part, the Newton–Thiele filter (NTF) instead of median filter is applied to remove impulse noise. The process to construct NTF can be divided into two steps: setting up the grid and constructing the Newton–Thiele's rational interpolation on the grid. First, eight adjacent pixels of the corrupted centre pixel are used to construct the two‐dimensional grid. If a pixel in the grid is corrupted, a four‐direction linear interpolation algorithm will be performed to provide a rough estimate to the corrupted pixel. Second, the corrupted centre pixel value will be updated by Newton–Thiele's rational interpolation on the grid. The NTF has better robustness than existing filters because it does not need to adjust window size or other parameters. Simulations reveal that the proposed detector and filter have perfect performance in terms of both quantitative evaluation and visual quality, especially it can remove the impulse noise effectively even at 90% noise level.