Neural surface reconstruction has been dominated by implicit representations with marching cubes for explicit surface extraction. However, those methods typically require high-quality normals for accurate reconstruction. We propose OffsetOPT, a method that reconstructs explicit surfaces directly from 3D point clouds and eliminates the need for point normals. The approach comprises two stages: first, we train a neural network to predict surface triangles based on local point geometry, given uniformly distributed training point clouds. Next, we apply the frozen network to reconstruct surfaces from unseen point clouds by optimizing a per-point offset to maximize the accuracy of triangle predictions. Compared to state-of-the-art methods, OffsetOPT not only excels at reconstructing overall surfaces but also significantly preserves sharp surface features. We demonstrate its accuracy on popular benchmarks, including small-scale shapes and large-scale open surfaces.

In the field of object detection, small object detection tasks have broad application prospects. However, detection models often face issues with insufficient image features for small objects and limited computational resources . To address these issues, we propose RC-SODet, a small object detector that uses reparameterization techniques combined with dual convolutions and compact feature enhancement blocks. In the detector, we design Reparameterized Dual Convolutions (RepDuConv) to replace conventional convolution and downsampling blocks. Its dual-branch advantage maintains accuracy, and the reparameterization technique built on this significantly improves inference efficiency. Compact Feature-enhanced Pyramid Network (RC-FPN) serves as the neck, using reparameterizable Cross Stage Partial with Feature Fusion Reparameterized Compact Blocks (C2fRCB) for feature enhancement. First, in the backbone network , RepDuConv replaces convolution blocks to perform downsampling on input images, thereby obtaining multi-scale features to pass to the neck. Second, the model uses RC-FPN as the feature pyramid neck to process multi-scale features from the backbone. After each front-end upsampling and fusion, dual-layer C2fRCB is applied to further refine and enhance the tensor features at different fusion scales. Finally, multi-level feature maps are fused at the back-end and passed to the detection head. Additionally, in the inference stage, both RepDuConv and C2fRCB optimize branch structures through reparameterization techniques. Experimental results show that on the small object datasets VisDrone and DroneVehicle, the highest version of RC-SODet achieves 48.1% and 82.4% mAP50, as well as 30.1% and 59.1% mAP50-95, respectively. The designed reparameterization technique increases the model inference speed by 58.1%.

Geometric feature learning for 3-D surfaces is critical for many applications in computer graphics and 3-D vision. However, deep learning currently lags in hierarchical modeling of 3-D surfaces due to the lack of required operations and/or their efficient implementations. In this article, we propose a series of modular operations for effective geometric feature learning from 3-D triangle meshes. These operations include novel mesh convolutions, efficient mesh decimation, and associated mesh (un)poolings. Our mesh convolutions exploit spherical harmonics as orthonormal bases to create continuous convolutional filters. The mesh decimation module is graphics processing unit (GPU)-accelerated and able to process batched meshes on-the-fly, while the (un)pooling operations compute features for upsampled/downsampled meshes. We provide an open-source implementation of these operations, collectively termed Picasso. Picasso supports heterogeneous mesh batching and processing. Leveraging its modular operations, we further contribute a novel hierarchical neural network for perceptual parsing of 3-D surfaces, named PicassoNet++. It achieves highly competitive performance for shape analysis and scene segmentation on prominent 3-D benchmarks. The code, data, and trained models are available at https://github.com/EnyaHermite/Picasso.

We present Picasso, a CUDA-based library comprising novel modules for deep learning over complex real-world 3D meshes. Hierarchical neural architectures have proved effective in multi-scale feature extraction which signifies the need for fast mesh decimation. However, existing methods rely on CPU-based implementations to obtain multi-resolution meshes. We design GPU-accelerated mesh decimation to facilitate network resolution reduction efficiently on-the-fly. Pooling and unpooling modules are defined on the vertex clusters gathered during decimation. For feature learning over meshes, Picasso contains three types of novel convolutions namely, facet2vertex, vertex2facet, and facet2facet convolution. Hence, it treats a mesh as a geometric structure comprising vertices and facets, rather than a spatial graph with edges as previous methods do. Picasso also incorporates a fuzzy mechanism in its filters for robustness to mesh sampling (vertex density). It exploits Gaussian mixtures to define fuzzy coefficients for the facet2vertex convolution, and barycentric interpolation to define the coefficients for the remaining two convolutions. In this release, we demonstrate the effectiveness of the proposed modules with competitive segmentation results on S3DIS. The library will be made public through github.

We propose a spherical kernel for efficient graph convolution of 3D point clouds. Our metric-based kernels systematically quantize the local 3D space to identify distinctive geometric relationships in the data. Similar to the regular grid CNN kernels, the spherical kernel maintains translation-invariance and asymmetry properties, where the former guarantees weight sharing among similar local structures in the data and the latter facilitates fine geometric learning. The proposed kernel is applied to graph neural networks without edge-dependent filter generation, making it computationally attractive for large point clouds. In our graph networks, each vertex is associated with a single point location and edges connect the neighborhood points within a defined range. The graph gets coarsened in the network with farthest point sampling. Analogous to the standard CNNs, we define pooling and unpooling operations for our network. We demonstrate the effectiveness of the proposed spherical kernel with graph neural networks for point cloud classification and semantic segmentation using ModelNet, ShapeNet, RueMonge2014, ScanNet and S3DIS datasets. The source code and the trained models can be downloaded from https://github.com/hlei-ziyan/SPH3D-GCN.

Fuzzy clustering is known to perform well in real-world applications. Inspired by this observation, we incorporate a fuzzy mechanism into discrete convolutional kernels for 3D point clouds as our first major contribution. The proposed fuzzy kernel is defined over a spherical volume that uses discrete bins. Discrete volumetric division can normally make a kernel vulnerable to boundary effects during learning as well as point density during inference. However, the proposed kernel remains robust to boundary conditions and point density due to the fuzzy mechanism. Our second major contribution comes as the proposal of an efficient graph convolutional network, SegGCN for segmenting point clouds. The proposed network exploits ResNet like blocks in the encoder and 1 × 1 convolutions in the decoder. SegGCN capitalizes on the separable convolution operation of the proposed fuzzy kernel for efficiency. We establish the effectiveness of the SegGCN with the proposed kernel on the challenging S3DIS and ScanNet real-world datasets. Our experiments demonstrate that the proposed network can segment over one million points per second with highly competitive performance.

We propose an octree guided neural network architecture and spherical convolutional kernel for machine learning from arbitrary 3D point clouds. The network architecture capitalizes on the sparse nature of irregular point clouds,and hierarchically coarsens the data representation with space partitioning. At the same time, the proposed spherical kernels systematically quantize point neighborhoods to identify local geometric structures in the data, while maintaining the properties of translation-invariance and asymmetry. We specify spherical kernels with the help of network neurons that in turn are associated with spatial locations.We exploit this association to avert dynamic kernel generation during network training that enables efficient learning with high resolution point clouds. The effectiveness of the proposed technique is established on the benchmark tasks of 3D object classification and segmentation, achieving competitive performance on ShapeNet and RueMonge2014 datasets.

In this paper, we present a robust global approach for point cloud registration from uniformly sampled points. Based on eigenvalues and normals computed from multiple scales, we design fast descriptors to extract local structures of these points. The eigenvalue-based descriptor is effective at finding seed matches with low precision using nearest neighbor search. Generally, recovering the transformation from matches with low precision is rather challenging. Therefore, we introduce a mechanism named correspondence propagation to aggregate each seed match into a set of numerous matches. With these sets of matches, multiple transformations between point clouds are computed. A quality function formulated from distance errors is used to identify the best transformation and fulfill a coarse alignment of the point clouds. Finally, we refine the alignment result with the trimmed iterative closest point algorithm. The proposed approach can be applied to register point clouds with significant or limited overlaps and small or large transformations. More encouragingly, it is rather efficient and very robust to noise. A comparison to traditional descriptor-based methods and other global algorithms demonstrates the fine performance of the proposed approach. We also show its promising application in large-scale reconstruction with the scans of two real scenes. In addition, the proposed approach can be used to register low-resolution point clouds captured by Kinect as well.

When blood flows through a bifurcation, red blood cells (RBCs) travel into side branches at different hematocrit levels, and it is even possible that all RBCs enter into one branch only, leading to a complete separation of plasma and RBCs. To quantify this phenomenon via particle-based mesoscopic simulations, we developed a general framework for open boundary conditions in multiphase flows that is effective even for high hematocrit levels. The inflow at the inlet is duplicated from a fully developed flow generated in a pilot simulation with periodic boundary conditions. The outflow is controlled by adaptive forces to maintain the flow rate and velocity gradient at fixed values, while the particles leaving the arteriole at the outlet are removed from the system. Upon validation of this approach, we performed systematic 3D simulations to study plasma skimming in arterioles of diameters 20 to 32 microns. For a flow rate ratio 6:1 at the branches, we observed the "all-or-nothing" phenomenon with plasma only entering the low flow rate branch. We then simulated blood-plasma separation in arteriolar bifurcations with different bifurcation angles and same diameter of the daughter branches. Our simulations predict a significant increase in RBC flux through the main daughter branch as the bifurcation angle is increased. Finally, we demonstrated the effectiveness of the new methodology in simulations of blood flow in vessels with multiple inlets and outlets, constructed using an angiogenesis model.

Red blood cells (RBCs) infected by a Plasmodium parasite in malaria may lose their membrane deformability with a relative membrane stiffening more than ten-fold in comparison with healthy RBCs leading to potential capillary occlusions. Moreover, infected RBCs are able to adhere to other healthy and parasitized cells and to the vascular endothelium resulting in a substantial disruption of normal blood circulation. In the present work, we simulate infected RBCs in malaria using a multiscale RBC model based on the dissipative particle dynamics method, coupling scales at the sub-cellular level with scales at the vessel size. Our objective is to conduct a full validation of the RBC model with a diverse set of experimental data, including temperature dependence, and to identify the limitations of this purely mechanistic model. The simulated elastic deformations of parasitized RBCs match those obtained in optical-tweezers experiments for different stages of intra-erythrocytic parasite development. The rheological properties of RBCs in malaria are compared with those obtained by optical magnetic twisting cytometry and by monitoring membrane fluctuations at room, physiological, and febrile temperatures. We also study the dynamics of infected RBCs in Poiseuille flow in comparison with healthy cells and present validated bulk viscosity predictions of malaria-infected blood for a wide range of parasitemia levels (percentage of infected RBCs with respect to the total number of cells in a unit volume).