Dynamic point cloud data, characterized by temporal continuity and high-dimensional features, can accurately capture dynamic changes of objects and are thus widely applied in fields such as autonomous driving, augmented reality, virtual reality, robotic navigation, and 3D video. Dynamic point cloud compression plays a crucial role in efficiently managing the storage, transmission, and perception of the ever-increasing volume of point cloud data. Despite this demand, there remains a clear lack of comprehensive surveys specifically focused on dynamic point cloud compression methods. To address this gap, this paper presents a systematic review of existing approaches, aiming to summarize recent advancements in the field and provide insights for future research. The paper first introduces the significance and theoretical foundations of dynamic point cloud compression. It then elaborates on the fundamental principles and advantages of mainstream models from three perspectives: geometry compression, attribute compression, and joint compression. Subsequently, commonly used datasets and evaluation metrics are reviewed, and the performance of various methods across different datasets is summarized. Finally, the paper analyzes the limitations of current methods in terms of algorithmic complexity and scene adaptability, and points out that lightweight network design, joint spatiotemporal modeling, and cross-modal information fusion are major directions for future research. This survey is expected to facilitate a deeper understanding of dynamic point cloud compression and promote further development of related systems.

Recent neural inverse rendering methods represent object geometry and materials with neural networks and learn network parameters from multi-view images through physically based rendering. However, these methods typically assume that the light source is located at an infinite distance, which seldomly holds in indoor scenarios that exhibit complex illumination. To address this issue, we propose a point cloud-based lighting representation method to model the spatially-varying and high-frequency lighting effects in indoor scenes. Our method first detects 2D light source masks in the input multi-view images, and then obtains a set of emitters through 3D reconstruction algorithm. We explicitly incorporate emitters in the Monte Carlo sampling, which improves the ability to model the specular effects, thus effectively alleviating the ambiguity in the inverse rendering process. Experiments on real and synthetic datasets demonstrate that the proposed method achieves the best performance in inverse rendering and can produce realistic relighting results.

The neural radiance field (NeRF), endowed with the implicit representation capability to model 3D scenes as continuous functions, facilitates the generation model in accomplishing 3D-aware image synthesis tasks on unstructured 2D image sets. Nevertheless, current approaches face challenges in image synthesis, including low fidelity, poor consistency and uncontrollability, stemming from their inability to fully capture the three-dimensional essence of the real world or their reliance on insufficiently expressive network architectures. To address these limitations, we introduce a locally spatial neural representation method grounded in a self-attention mechanism. This approach augments the NeRF generator, empowering it to achieve controllable high-quality 3D-aware image synthesis. Initially, the scene to be synthesized is decomposed by incorporating the pose information of each object and the predefined 3D boundary box. Subsequently, the self-attention local representation module models each object distinctly. A defined combination operator is then utilized to obtain the complete scene representation. Finally, the rendering module processes the resulting RGB image. Experimental results show that the proposed method accomplishes the task of high-quality 3D-aware image synthesis with multi-view consistency and controllability, even under conditions of low-frequency spatial point sampling. Furthermore, when conducting synthesis testing with resolutions of 64×64 and 256×256 pixels on the highly complex CompCars public dataset, our proposed method achieved a 4.69 and 2.51 improvement in terms of FID scores compared to the baseline method, respectively. Similarly, on the Churches public dataset, the FID scores were improved by 1.93 and 3.56, respectively, further highlighting the effectiveness of the method.

Subsurface scattering is a key factor in ensuring the visual effects of translucent materials, and it holds significant application value in film, gaming, virtual reality, and digital twin technology. However, existing technologies rely on planar diffusion profiles, which lead to issues such as color distortion and harsh shadow transitions when rendering complex surfaces. To address these issues, this paper proposes a pre-integration method based on multipole theory to enhance the accuracy of subsurface scattering on surface models. Initially, it employs the multipole model to build accurate single-layer and multi-layer diffusion profiles for surfaces. Subsequently, these diffusion profiles are pre-integrated into a Lookup Table (LUT), and single scattering is included to further improve the scattering precision. Lastly, during real-time rendering, texture blur techniques are used in conjunction with LUT to improve the accuracy of the average curvature. Moreover, the paper discusses the contribution of transmitted light to the scattering effect in order to enhance the scattering in specific scenarios. Experiments focused on facial rendering demonstrate that the proposed method outperforms existing pre-integration techniques in error metrics such as SSIM and RMSE, and effectively eliminates previous errors in surface scattering while maintaining rendering speed and significantly enhancing the realism of the render.

To solve the problem of feature distortion induced by excessive smoothing or sharpening during the reconstruction of 3D point cloud, a feature-preserving probabilistic reconstruction method is proposed. First, low-rank estimation of local curvature of point cloud is realized by space-based principal component analysis in order to better locate sharp features, and the normal field of point cloud is adaptively optimized based on curvature which guarantees accurate normal constraints during sharp feature recovery. Then, the probability reconstruction model of point cloud is established according to Bayesian statistic inference. And the posterior probability distribution of each point position is simulated by sequential Monte Carlo method within particle filtering framework. In order to improve the convergence, the resampling range is adjusted according to the local noise. Finally, the Bayesian reconstruction of point cloud is achieved using resampling with normal and spherical constraints for different regions which takes the advantage of multiple hypothesis testing, and the problems of feature degradation and irregular edges are solved. Compared with several state-of-the-art methods, the experimental results show that the proposed algorithm has improved subjective visual effects, reconstruction accuracy, and efficiency, and the reconstruction error is reduced by 46.84% while the running time is decreased by 33.34% on average. The proposed method can effectively restore original features while suppressing noise, which improve the quality of point cloud.

Sketch-based 3D model retrieval is an important way to retrieve 3D models in mechanical part library. There are huge modal differences between sketches and 3D models. Aiming at the information redundancy of multiple views projected by the 3D model, the weak representation ability of the features extracted from the part views, in order to improve the effect of sketch-based 3D model retrieval, a sketch-based 3D mechanical part Library retrieval method is proposed to fuse the local and global dual-scale features of the part model. Firstly, a projection view representation method of 3D model based on image entropy is proposed to reduce the redundancy among multiple views. Then, a mechanical sketch model for edge contour extraction of part view is proposed, and the contour sketch of the part is extracted in depth to approximate the similarity between the view and the sketch. Finally, a visual word bag model and an improved multi-view convolutional neural network model is constructed to extract the local and global features of the contour sketch respectively. By fusing the dual-scale features, Euclidean distance is used to match the 3D model of the part based on the sketch. The experimental results on ESB datasets show that the proposed method is significantly improved in a series of evaluation indicators such as accuracy and recall rate compared with the existing 6 common methods, which verifies the feasibility and effectiveness of the proposed method.

Topic modelling is an important text mining method in natural language processing, but its modeling process is complex and can generate results that do not always align with user expectations. To address this issue and enable non-expert users to understand the modelling process and modify model results quickly, we propose the interactive visual analysis system BTVis based on BERTopic. The system enhances the interpretability and usability of BERTopic through the following key features: (1) explored and visualized the intermediate process of hierarchical clustering of BERTopic to intuitively reveal the topic generation mechanism; (2) analyzed the outlier documents to uncover potential relationships between the outlier documents and the topics; (3) proposed the multi-granular local model editing algorithms to enhance the accuracy of the BERTopic model; (4) implemented the interactive hierarchical topic modeling system BTVis in a web environment, allowing users to enhance BERTopic results through visual analysis and interactive exploration. We conducted experiments including qualitative analysis, user experiments and quantitative tests with multiple real datasets. The results show the effectiveness and practicality of the system.

Existing approaches to semantic pattern exploration limit the comprehension of global semantic patterns, with few studies focusing on the fine-grained semantic pattern exploration that preserves contextual semantics. In this paper, this paper combines machine learning and visualization techniques in depth, employing a fine-grained semantic pattern visual analysis method for multi-attribute documents. This allows users to flexibly analyze fine-grained semantic patterns under different attributes while perceiving global semantic patterns. Firstly, this paper introduces a representation reconstruction network to obtain latent vectors that contain semantic and attribute information of multi-attribute documents, enabling the topic model to better identify subspace topics. Then, this paper introduces a human-in-the-loop semantic pattern visual analysis method and develop a semantic pattern visualization system that includes an exploration manager, a subspace projector, and a topic interpreter. This system supports users in selecting attribute subspaces for representation reconstruction, interactively exploring semantic patterns, and providing analysis results. Based on game review datasets, US news datasets, and Trump election datasets, this paper compare our proposed topic model with existing methods using topic diversity and topic coherence metrics. Experimental results show that the proposed topic model has good generalizability and flexibility in topic modeling. User experiments including semantic pattern exploration tasks verify the outstanding efficiency and effectiveness of the proposed method and visualization system in semantic pattern exploration.

Existing learning-based feature point extraction methods neglect long-range context information. Thus, existing learning-based feature point extraction methods perform poorly in repetitive-texture and low-texture scenes such as oil production scenes. This problem poses difficulties for the analysis of safety video surveillance and the construction of digital twin platforms in oil field production. To solve this problem, we propose a novel and efficient feature point extraction network that fuses local and long-range information. Firstly, the proposed network utilizes efficient large-kernel convolutional layers to capture long-range information. Then, the proposed network can extract local descriptors with higher discriminative ability by fusing the long-range information and the local information captured by vanilla convolutional layers. To further enhance the matching accuracy of feature points, we propose a novel method of adaptively constructing hardest descriptor triplets within entire images. The proposed method extends the scope of constructing triplets from covisible areas to entire images, which raises the similarity between the training process and real application scenarios. On the HPatches dataset, the mean matching accuracy of the proposed method is 2.89% higher than those of the comparison methods. On the Aachen Day-Night v1.1 dataset, the visual localization accuracy of the proposed method is 2.10% higher than those of the comparison methods. On the InLoc dataset, the visual localization accuracy of the proposed method is 3.70% higher than those of the comparison methods. On the dataset of oil production scenes with repetitive and weak texture regions, the average number of feature point matching inliers of the proposed method is 20.37% higher than those of the comparison methods.

In order to solve the problems that existing image dehazing methods cannot effectively remove non-homogeneous haze and their iterative dehazing results takes on overall darkening effects, a brightness compensation based iterative non-homogeneous image dehazing algorithm framework is proposed with the image dehazing module and the brightness compensation module. In each iteration, the image dehazing module is first employed to enhance the contrast of the image and reduce the haze density. Subsequently, the brightness compensation module is utilized to make up for the brightness loss caused by the image dehazing module, ensuring that the results of brightness compensation not only maintain the previous image dehazing effect but also achieve the original brightness level before the dehazing procedure. This process creates favorable conditions for further improving the dehazing effect of the image in the next iteration. In the brightness compensation module, a brightness compensation operator is designed according to the brightness changes before and after the image dehazing module, and a brightness compensation coefficients based adaptive brightness compensation mechanism is established, which can not only restore the original brightness of the image, but also improve the contrast of the image which can further boost the image dehazing effects. Extensive experimental results on the FiveK-HAZE, SOTS test set, and Real-Haze dataset demonstrate outstanding performance of the proposed algorithm in handling non-homogeneous haze images. The dehazed results exhibit excellent brightness, contrast, and color saturation. Our algorithm outperforms comparison methods in commonly used evaluation metrics such as peak signal-to-noise ratio, structural similarity, haze density, image naturalness, and learned perceptual image patch similarity.

Sharing videos on the Internet is becoming increasingly common, therefore, video quality assessment is becoming increasingly important. In existing video quality assessment research, convolution is widely used for feature extraction in the spatial and temporal domains of video, but due to its inherent local perceptual characteristics, the global information perception for video is neglected. At the same time, existing methods do not fully integrate spatial and temporal features. To solve the above problems, this paper proposes a no-reference video quality assessment method that combines convolution with a self-attention mechanism and spatiotemporal fusion. To extract more effective spatial and temporal features, the method first designs a spatial feature extraction module and a temporal feature extraction module, where the spatial feature extraction module enhances the global information of the local features extracted by convolution through long-range dependencies by the proposed feature perception and enhancement module that combines convolution and self-attention. The temporal feature extraction module jointly learns the temporal features of video using Video Swin Transformer and 3D convolution based on video frame sequences and video frame difference sequences. Then, the spatiotemporal feature fusion and enhancement module is used to fuse and enhance the temporal and spatial features through the self-attention mechanism, and finally predict the video quality. Experimental results on the KoNVid-1k and YouTube-UGC datasets show that the proposed method has better performance compared to existing video quality assessment methods, and ablation experiments also verify the effectiveness of each proposed module.

To address the issue of existing deep learning-based segmentation methods for dental models, which overly rely on coordinate features and overlook the differences and complementary nature of geometric and coordinate features, a segmentation network based on attention mechanism and neighbor enhancement is proposed. First, an attention-enhanced module is employed to enhance the neighborhood of input features through different encoding schemes. It designs a differential attention mechanism and a self-attention mechanism to extract comprehensive contextual features of the mesh by considering the properties of geometric and coordinate features separately. Next, a complementary pooling function is constructed to fuse maximum pooling, attention pooling, and average pooling, which promotes the complementary interaction among different features and enhances the robustness of the segmentation network. Finally, an adaptive attention module is embedded to balance the numerical differences among multi-scale features for effective mesh segmentation. Compared with other methods in 200 real patients dental model dataset, the segmentation accuracy of dental model is 96.18%. The experimental results show that the proposed method is superior to the compared deep learning networks and has better performance in the segmentation of dental models.

Existing few-shot Chinese font generation models struggle to effectively decouple font style and content using a limited number of reference characters, leading to suboptimal font generation results. To address this issue, a novel few-shot Chinese font generation method based on contrastive learning and square-block transformation data augmentation is proposed. Utilizing the concept of contrastive learning, the output images undergo separat e style and content comparisons, achieving better adaptive decoupling of font style and content. To further improve the performance of few-shot font generation, a data augmentation method based on square-block transformations is introduced, embedding the global structural information of the fonts through geometric transformations of the square-block areas, thereby enhancing the ability to decouple font style and content. The effectiveness of the proposed method is validated through extensive experiments on public Chinese font datasets. Experimental results demonstrate that, compared to current state-of-the-art few-shot font generation models, the proposed method significantly improves the quality of generated fonts and outperforms comparative models in performance metrics such as FID, SSIM, PSNR, and LPIPS. Notably, the proposed method can be integrated into existing few-shot font generation models and, with only a small training dataset, can significantly enhance the performance of these models.

Interpretable visual image representation learning to reveal image variation factors is a hot research topic in computer vision. Many existing disentanglement methods discover variation factors of images and learn disentangled representations by using extra regularization term. However, it usually leads to an imbalance between disentanglement and generative quality, which affects visual image understanding. To address this issue, a generative visual image understanding method based on disentangled representation learning is proposed in terms of interpretable variations in images. Firstly, a pre-trained Glow model is designed to acquire the latent representations of target images. Secondly, a learning strategy based on image variation is constructed from the latent representations to obtain interpretable directions of candidate traversals. Finally, the contrast module is designed under the contrastive learning perspective to simulate image variations based on the interpretable directions of candidate traversals and then extract disentangled representations. The experimental results show that better results are achieved on the popular disentanglement datasets, which are Shapes3D, MPI3D, Anime, MNIST and Cars3D, where the MIG, DCI, FactorVAE score and β-VAE score metrics reach 0.16, 0.27, 0.89 and 0.98, respectively, on the Cars3D dataset, verifying the effectiveness and feasibility of the proposed method.

To solve the problems of blurring and low resolution in depth maps captured by low-cost depth cameras, a hybrid multimode attention fusion correction network is proposed by combining high-resolution color images in the same scene. To effectively leverage the structure information of color images, the attention mechanism is employed to iteratively align and integrate depth features with color structural features, and the fused features are corrected through the incorporation of color structure features and edge aware maps. Initially, the deep edge aware block combines low-resolution depth maps with color edge maps to extract high-resolution depth edge maps. Then, the side-window edge preserve block is utilized to extract useful structural information from the color image, reducing the impact of extraneous texture information. Finally, multimodal attention fusion correction block utilizes attention mechanisms to extract the structural attention distribution of both depth features and color structural features, aligning and fusing these modalities. The spatial attention weights of the color structural features and high-resolution depth edge maps are used to enhance the supervision of the depth edges, reducing blurring and artifacts. Experimental results on the NYU v2 dataset and Middlebury dataset (2005) show that the average root mean square error test results are 2.96 and 1.45, respectively, which are reduced by 0.10 and 0.08 compared to suboptimal methods. The network can reconstruct high-resolution depth maps with clearer edges and fewer artifacts.