This paper presents EVSplitting, an efficient and visually consistent splitting algorithm for 3D Gaussian Splatting (3DGS). It is designed to make operating 3DGS as easy and effective as other 3D explicit representations, readily for industrial productions. The challenges of above target are: 1) The huge number and complex attributes of 3DGS make it tough to explicitly operate on 3DGS in a real-time and learning-free manner; 2) The visual effect of 3DGS is very difficult to maintain during explicit operations and 3) The anisotropism of Gaussian always leads to blurs and artifacts. As far as we know, no prior work can address these challenges well. In this work, we introduce a direct and efficient 3DGS splitting algorithm to solve them. Specifically, we formulate the 3DGS splitting as two minimization problems that aim to ensure visual consistency and reduce Gaussian overflow across boundary (splitting plane), respectively. Firstly, we impose conservations on the zero-, first- and second-order moments of the weighted Gaussian distribution to guarantee visual consistency. Secondly, we reduce the boundary overflow with a special constraint on the aforementioned conservations. With these conservations and constraints, we derive a closed-form solution for the 3DGS splitting problem. This yields an easy-to-implement, plug-and-play, efficient and fundamental tool, benefiting various downstream applications of 3DGS.

3D Gaussian splatting models, as a novel explicit 3D representation, have been applied in many domains recently, such as explicit geometric editing and geometry generation. Progress has been rapid. However, due to their mixed scales and cluttered shapes, 3D Gaussian splatting models can produce a blurred or needle-like effect near the surface. At the same time, 3D Gaussian splatting models tend to flatten large untextured regions, yielding a very sparse point cloud. These problems are caused by the non-uniform nature of 3D Gaussian splatting models, so in this paper, we propose a new 3D Gaussian splitting algorithm, which can produce a more uniform and surface-bounded 3D Gaussian splatting model. Our algorithm splits an N-dimensional Gaussian into two N-dimensional Gaussians. It ensures consistency of mathematical characteristics and similarity of appearance, allowing resulting 3D Gaussian splatting models to be more uniform and a better fit to the underlying surface, and thus more suitable for explicit editing, point cloud extraction and other tasks. Meanwhile, our 3D Gaussian splitting approach has a very simple closed-form solution, making it readily applicable to any 3D Gaussian model.

Deep neural networks (DNNs) have been widely used for mesh processing in recent years. However, current DNNs can not process arbitrary meshes efficiently. On the one hand, most DNNs expect 2-manifold, watertight meshes, but many meshes, whether manually designed or automatically generated, may have gaps, non-manifold geometry, or other defects. On the other hand, the irregular structure of meshes also brings challenges to building hierarchical structures and aggregating local geometric information, which is critical to conduct DNNs. In this paper, we present DGNet, an efficient, effective and generic deep neural mesh processing network based on dual graph pyramids; it can handle arbitrary meshes. First, we construct dual graph pyramids for meshes to guide feature propagation between hierarchical levels for both downsampling and upsampling. Second, we propose a novel convolution to aggregate local features on the proposed hierarchical graphs. By utilizing both geodesic neighbors and euclidean neighbors, the network enables feature aggregation both within local surface patches and between isolated mesh components. Experimental results demonstrate that DGNet can be applied to both shape analysis and large-scale scene understanding. Furthermore, it achieves superior performance on various benchmarks, including ShapeNetCore, HumanBody, ScanNet and Matterport3D. Code and models will be available at https://github.com/li-xl/DGNet.

Inspired by the success of recent vision transformers and large kernel design in convolutional neural networks (CNNs), in this paper, we analyze and explore essential reasons for their success. We claim two factors that are critical for 3D large-scale scene understanding: a larger receptive field and operations with greater non-linearity. The former is responsible for providing long range contexts and the latter can enhance the capacity of the network. To achieve the above properties, we propose a simple yet effective long range pooling (LRP) module using dilation max pooling, which provides a network with a large adaptive receptive field. LRP has few parameters, and can be readily added to current CNNs. Also, based on LRP, we present an entire network architecture, LRPNet, for 3D understanding. Ablation studies are presented to support our claims, and show that the LRP module achieves better results than large kernel convolution yet with reduced computation, due to its non-linearity. We also demonstrate the superiority of LRPNet on various benchmarks: LRPNet performs the best on ScanNet and surpasses other CNN-based methods on S3DIS and Matterport3D. Code will be avalible at https://github.com/li-xl/LRPNet.

Objects in aerial images show greater variations in scale and orientation than in other images, making them harder to detect using vanilla deep convolutional neural networks. Networks with sampling equivariance can adapt sampling from input feature maps to object transformation, allowing a convolutional kernel to extract effective object features under different transformations. However, methods such as deformable convolutional networks can only provide sampling equivariance under certain circumstances, as they sample by location. We propose sampling equivariant self-attention networks, which treat self-attention restricted to a local image patch as convolution sampling by masks instead of locations, and a transformation embedding module to improve the equivariant sampling further. We further propose a novel randomized normalization module to enhance network generalization and a quantitative evaluation metric to fairly evaluate the ability of sampling equivariance of different models. Experiments show that our model provides significantly better sampling equivariance than existing methods without additional supervision and can thus extract more effective image features. Our model achieves state-of-the-art results on the DOTA-v1.0, DOTA-v1.5, and HRSC2016 datasets without additional computations or parameters.

Humans can naturally and effectively find salient regions in complex scenes. Motivated by thisobservation, attention mechanisms were introduced into computer vision with the aim of imitating this aspect of the human visual system. Such an attention mechanism can be regarded as a dynamic weight adjustment process based on features of the input image. Attention mechanisms have achieved great success in many visual tasks, including image classification, object detection, semantic segmentation, video understanding, image generation, 3D vision, multi-modal tasks, and self-supervised learning. In this survey, we provide a comprehensive review of various attention mechanisms in computer vision and categorize them according to approach, such as channel attention, spatial attention, temporal attention, and branch attention; a related repository https://github.com/MenghaoGuo/Awesome-Vision-Attentions is dedicated to collecting related work. We also suggest future directions for attention mechanism research.

Our goal is to automatically generate CAD 3D models from 2D sketches as part of a design chain where models should be procedural, containing features arranged in a model tree and linked to suitable datums. Current procedural models capture much about the design intent and are easy to edit, but must be created from scratch during the detailed design state-given conceptual sketches as used by designers in the early part of the design process, current sketch-based modeling approaches only output explicit models. Thus, we describe an approach to extract high-level information directly from 2D engineering wireframe sketches and use it to complete a CSG feature tree, which serves as a model tree for a procedural 3D CAD model. Our method extracts procedural model information directly from 2D sketches in the form of a set of features, plus a set of datums and relationships between these features. We detect and analyze features of 2D sketches in isolation, and define the CSG feature tree by the parent-child relationships between features, and combine this information to obtain a complete and consistent CSG feature tree that can be transferred to a 3D modeler, which reconstructs the model. This paper focuses on how to extract the feature datums and the extrusion operation from an input 2D sketch. (c) 2021 The Authors. Published by Elsevier Ltd.

Convolutionalneural networks (CNNs) have made great breakthroughs in two-dimensional (2D) computer vision. However, their irregular structure makes it hard to harness the potential of CNNs directly on meshes. A subdivision surface provides a hierarchical multi-resolution structure in which each face in a closed 2-manifold triangle mesh is exactly adjacent to three faces. Motivated by these two observations, this article presents SubdivNet, an innovative and versatile CNN framework for three-dimensional (3D) triangle meshes with Loop subdivision sequence connectivity. Making an analogy between mesh faces and pixels in a 2D image allows us to present a mesh convolution operator to aggregate local features from nearby faces. By exploiting face neighborhoods, this convolution can support standard 2D convolutional network concepts, e.g., variable kernel size, stride, and dilation. Based on the multi-resolution hierarchy, we make use of pooling layers that uniformly merge four faces into one and an upsampling method that splits one face into four. Thereby, many popular 2D CNN architectures can be easily adapted to process 3D meshes. Meshes with arbitrary connectivity can be remeshed to have Loop subdivision sequence connectivity via self-parameterization, making SubdivNet a general approach. Extensive evaluation and various applications demonstrate SubdivNet’s effectiveness and efficiency.

Joint pose estimation of human hands and objects from a single RGB image is an important topic for AR/VR, robot manipulation, etc. It is common practice to determine both poses directly from the image; some recent methods attempt to improve the initial poses using a variety of contact-based approaches. However, few methods take the real physical constraints conveyed by the image into consideration, leading to less realistic results than the initial estimates. To overcome this problem, we make use of a set of high-level 2D features which can be directly extracted from the image in a new pipeline which combines contact approaches and these constraints during optimization. Our pipeline achieves better results than direct regression or contact-based optimization: they are closer to the ground truth and provide high quality contact.

Accurate camera pose estimation is essential and challenging for real world dynamic 3D reconstruction and augmented reality applications. In this article, we present a novel RGB-D SLAM approach for accurate camera pose tracking in dynamic environments. Previous methods detect dynamic components only across a short time-span of consecutive frames. Instead, we provide a more accurate dynamic 3D landmark detection method, followed by the use of long-term consistency via conditional random fields, which leverages long-term observations from multiple frames. Specifically, we first introduce an efficient initial camera pose estimation method based on distinguishing dynamic from static points using graph-cut RANSAC. These static/dynamic labels are used as priors for the unary potential in the conditional random fields, which further improves the accuracy of dynamic 3D landmark detection. Evaluation using the TUM and Bonn RGB-D dynamic datasets shows that our approach significantly outperforms state-of-the-art methods, providing much more accurate camera trajectory estimation in a variety of highly dynamic environments. We also show that dynamic 3D reconstruction can benefit from the camera poses estimated by our RGB-D SLAM approach.

The irregular domain and lack of ordering make it challenging to design deep neural networks for point cloud processing. This paper presents a novel framework named Point Cloud Transformer (PCT) for point cloud learning. PCT is based on Transformer, which achieves huge success in natural language processing and displays great potential in image processing. It is inherently permutation invariant for processing a sequence of points, making it well-suited for point cloud learning. To better capture local context within the point cloud, we enhance input embedding with the support of farthest point sampling and nearest neighbor search. Extensive experiments demonstrate that the PCT achieves the state-of-the-art performance on shape classification, part segmentation, semantic segmentation, and normal estimation tasks.

Understanding how people perceive the visual complexity of shapes has important theoretical as well as practical implications. One school of thought, driven by information theory, focuses on studying the local features that contribute to the perception of visual complexity. Another school, in contrast, emphasizes the impact of global characteristics of shapes on perceived complexity. Inspired by recent discoveries in neuroscience, our model considers both local features of shapes: edge lengths and vertex angles, and global features: concaveness, and is in 92% agreement with human subjective ratings of shape complexity. The model is also consistent with the hierarchical perceptual learning theory, which explains how different layers of neurons in the visual system act together to yield a perception of visual shape complexity.

Rigid registration is the process of identifying the correspondences between two overlapping shapes that were subject to a rigid transformation. Once the correspondences have been established, the underlying rigid transformation can be estimated in the weighted least squares sense with closed form solution. A correspondence is such a pair of points that while they are described in two different coordinate systems, they represent the same point on the object surface in 3D space.

In the first week of May, 2021, researchers from four different institutions: Google, Tsinghua University, Oxford University and Facebook, shared their latest work [16, 7, 12, 17] on arXiv.org almost at the same time, each proposing new learning architectures, consisting mainly of linear layers, claiming them to be comparable, or even superior to convolutional-based models. This sparked immediate discussion and debate in both academic and industrial communities as to whether MLPs are sufficient, many thinking that learning architectures are returning to MLPs. Is this true? In this perspective, we give a brief history of learning architectures, including multilayer perceptrons (MLPs), convolutional neural networks (CNNs) and transformers. We then examine what the four newly proposed architectures have in common. Finally, we give our views on challenges and directions for new learning architectures, hoping to inspire future research.

The paper proposes a general numerical framework to simplify a CAD model into a volume mesh model under reliable control of certain prescribed physical quantity that the designer is interested in. Different from previous work, the proposed approach does not assume that the candidate features have been detected and can directly generate the simplified volume mesh model. In addition, it can efficiently estimate the quantitative impact of each individual feature via solving a linear equation of small dimension less than 10. This is achieved by reformulating the problem as estimating the solution differences caused by different stiffness matrices, using the combined approximation approach. Performance of this approach is demonstrated via numerical 2D examples.

To realistically represent 3D human body shape in a mathematical way, the parametric model used should incorporate symmetry as displayed by real people. This paper proposes a symmetric parametric model called symmetricSCAPE. It successfully incorporates symmetry into a parametric model of the 3D body, formulating body geometric variations of both shape and pose using a triangular mesh representation. The symmetry constraint is imposed on each symmetrically-related triangle pair of the body mesh. Mathematically, symmetry-related constraint matrices are derived, and applied during shape and pose deformation of triangle pairs. By accurately registering a pre-designed symmetrization template mesh to the training dataset, we learn how the symmetricSCAPE model causes the body mesh to deform relative to the symmetry. Our experiments demonstrate that the symmetricSCAPE model results in a better, more parsimonious, and more accurate parametric model of the 3D human body than traditional non symmetry -aware representations. (C) 2019 Elsevier Ltd. All rights reserved.

In order to accurately count the number of animals grazing on grassland, we present a livestock detection algorithm using modified versions of U-net and Google Inception-v4 net. This method works well to detect dense and touching instances. We also introduce a dataset for livestock detection in aerial images, consisting of 89 aerial images collected by quadcopter. Each image has resolution of about 3000×4000 pixels, and contains livestock with varying shapes, scales, and orientations. We evaluate our method by comparison against Faster RCNN and Yolo-v3 algorithms using our aerial livestock dataset. The average precision of our method is better than Yolo-v3 and is comparable to Faster RCNN.

A quick-response code (QR code) is a two-dimensional code akin to a barcode that encodes a message of limited length. In this paper, we present a variant of QR code, a two-layer QR code. Its two-layer structure can display two alternative messages when scanned from two different directions. We propose a method to generate such two-layer QR codes encoding two given messages in a few seconds. We also demonstrate the robustness of our method on both synthetic and fabricated examples. All source code will be made publicly available (https://github.com/yuantailing/two-layer-qrcode).