Weakly Supervised Part-wise 3D Shape Reconstruction from Single-View RGB Images

In order for the deep learning models to truly understand the 2D images for 3D geometry recovery, we argue that single-view reconstruction should be learned in a part-aware and weakly supervised manner. Such models lead to more profound interpretation of 2D images in which part-based parsing and assembling are involved. To this end, we learn a deep neural network which takes a single-view RGB image as input, and outputs a 3D shape in parts represented by 3D point clouds with an array of 3D part generators. In particular, we devise two levels of generative adversarial network (GAN) to generate shapes with both correct part shape and reasonable overall structure. To enable a self-taught network training, we devise a differentiable projection module along with a self-projection loss measuring the error between the shape projection and the input image. The training data in our method is unpaired between the 2D images and the 3D shapes with part decomposition. Through qualitative and quantitative evaluations on public datasets, we show that our method achieves good performance in part-wise single-view reconstruction.     

自监督深度残差函数映射网络的3维模型对应关系计算

目的 针对传统非刚性3维模型的对应关系计算方法需要模型间真实对应关系监督的缺点,提出一种自监督深度残差函数映射网络（self-supervised deep residual functional maps network,SSDRFMN）。方法 首先将局部坐标系与直方图结合以计算3维模型的特征描述符,即方向直方图签名（signature of histograms of orientations,SHOT）描述符;其次将源模型与目标模型的SHOT描述符输入SSDRFMN,利用深度函数映射（deep functional maps,DFM）层计算两个模型间的函数映射矩阵,并通过模糊对应层将函数映射关系转换为点到点的对应关系;最后利用自监督损失函数计算模型间的测地距离误差,对计算出的对应关系进行评估。结果 实验结果表明,在MPI-FAUST数据集上,本文算法相比于有监督的深度函数映射（supervised deep functional maps,SDFM）算法,人体模型对应关系的测地误差减小了1.45;相比于频谱上采样（spectral upsampling,SU）算法减小了1.67。在TOSCA数据集上,本文算法相比于SDFM算法,狗、猫和狼等模型的对应关系的测地误差分别减小了3.13、0.98和1.89;相比于SU算法分别减小了2.81、2.22和1.11,并有效克服了已有深度函数映射方法需要模型间的真实对应关系来监督的缺点,使得该方法可以适用于不同的数据集,可扩展性大幅增强。结论 本文通过自监督深度残差函数映射网络训练模型的方向直方图签名描述符,提升了模型对应关系的准确率。本文方法可以适应于不同的数据集,相比传统方法,普适性较好。     

Dirichlet Harmonic Shape Compression with Feature Preservation for Parameterized Surfaces

With the rapid advancement of 3D scanning devices, large and complicated 3D shapes are becoming ubiquitous, and require large amount of resources to store and transmit them efficiently. This makes shape compression a demanding technique in order for the user to reduce the data transmission latency. Existing shape compression methods could achieve very low bit‐rates by sacrificing shape quality. But none of them guarantees the preservation of salient feature lines that users care. In addition, many 3D shapes come with parametric information for texture mapping purposes. In this paper we describe a spectral method to compress the geometric shapes equipped with arbitrary valid parametric information. It guarantees to preserve user‐specified feature lines while achieving a high compression ratio. By applying the spectral shape analysis – Dirichlet Manifold Harmonics, in the 2D parametric domain, this method provides a progressive compression mechanism to trade‐off between bit‐rate and shape quality. Experiments show that this method provides very low bit‐rate with high shape‐quality and still guarantees the preservation of user‐specified feature lines.     

Learning Generative Models of 3D Structures

3D models of objects and scenes are critical to many academic disciplines and industrial applications. Of particular interest is the emerging opportunity for 3D graphics to serve artificial intelligence: computer vision systems can benefit from synthetically-generated training data rendered from virtual 3D scenes, and robots can be trained to navigate in and interact with real-world environments by first acquiring skills in simulated ones. One of the most promising ways to achieve this is by learning and applying generative models of 3D content: computer programs that can synthesize new 3D shapes and scenes. To allow users to edit and manipulate the synthesized 3D content to achieve their goals, the generative model should also be structure-aware: it should express 3D shapes and scenes using abstractions that allow manipulation of their high-level structure. This state-of-the-art report surveys historical work and recent progress on learning structure-aware generative models of 3D shapes and scenes. We present fundamental representations of 3D shape and scene geometry and structures, describe prominent methodologies including probabilistic models, deep generative models, program synthesis, and neural networks for structured data, and cover many recent methods for structure-aware synthesis of 3D shapes and indoor scenes.     

基于多任务学习的有限样本多视角三维形状识别算法

随着三维扫描技术的快速发展,三维形状分析得到了学术界的广泛关注;尤其是深度学习在计算机视觉上取得的显著成功,使得基于多视图的三维形状识别方法成为了目前三维模型识别的主流方式。已有研究表明,三维数据集的数量对于最终的分类精度是一个非常重要的影响条件。然而,由于专业三维扫描设备的限制,三维形状数据难以采集。实际上,现有的公共基准三维数据集的规模远远小于二维数据集,三维形状分析的发展因此受到阻碍。为了解决这一问题,文中主要研究在极小数据样本情况下,三维形状识别问题的优化解策略。受多任务学习的启发,搭建了多分支的网络结构,并引入基于度量学习的辅助比较模块,用于挖掘类内和类间的相似性和差异性信息。网络模型包括主支路与辅助支路,分别使用不同的损失函数对应不同的训练任务,并使用权值超参数平衡多项损失。主支路获得预测分类,使用交叉熵损失函数进行更新;辅助支路得到不同样本间的相似性得分,使用均方差损失函数进行更新。为保证特征向量被投影到同一个空间中,主、辅助支路共享相同的特征提取模块,在训练阶段共同更新参数,在测试阶段仅使用主支路获得的分类结果。在两个公开的三维形状基准数据集上的大量实验结果表明,所提网络结构与训练策略相比传统方法,在少样本的情况下可以显著提高特征模块对不同类别的区分能力,获得更优的识别结果。     

Semi-Supervised 3D Shape Recognition via Multimodal Deep Co-training

3D shape recognition has been actively investigated in the field of computer graphics. With the rapid development of deep learning, various deep models have been introduced and achieved remarkable results. Most 3D shape recognition methods are supervised and learn only from the large amount of labeled shapes. However, it is expensive and time consuming to obtain such a large training set. In contrast to these methods, this paper studies a semi-supervised learning framework to train a deep model for 3D shape recognition by using both labeled and unlabeled shapes. Inspired by the co-training algorithm, our method iterates between model training and pseudo-label generation phases. In the model training phase, we train two deep networks based on the point cloud and multi-view representation simultaneously. In the pseudo-label generation phase, we generate the pseudo-labels of the unlabeled shapes using the joint prediction of two networks, which augments the labeled set for the next iteration. To extract more reliable consensus information from multiple representations, we propose an uncertainty-aware consistency loss function to combine the two networks into a multimodal network. This not only encourages the two networks to give similar predictions on the unlabeled set, but also eliminates the negative influence of the large performance gap between the two networks. Experiments on the benchmark ModelNet40 demonstrate that, with only 10% labeled training data, our approach achieves competitive performance to the results reported by supervised methods.     

Data‐Driven Shape Analysis and Processing

Data‐driven methods serve an increasingly important role in discovering geometric, structural and semantic relationships between shapes. In contrast to traditional approaches that process shapes in isolation of each other, data‐driven methods aggregate information from 3D model collections to improve the analysis, modelling and editing of shapes. Data‐driven methods are also able to learn computational models that reason about properties and relationships of shapes without relying on hard‐coded rules or explicitly programmed instructions. Through reviewing the literature, we provide an overview of the main concepts and components of these methods, as well as discuss their application to classification, segmentation, matching, reconstruction, modelling and exploration, as well as scene analysis and synthesis. We conclude our report with ideas that can inspire future research in data‐driven shape analysis and processing.     

利用隐式解码器的三维模型簇协同分割

目的为建立3维模型语义部件之间的对应关系并实现模型自动分割,提出一种利用隐式解码器(implicit decoder, IM-decoder)的无监督3维模型簇协同分割网络。方法首先对3维点云模型进行体素化操作,进而由CNN-encoder(convolutional neural network encoder)提取体素化点云模型的特征,并将模型信息映射至特征空间。然后使用注意力模块聚合3维模型相邻点特征,将聚合特征与3维点坐标作为IM-decoder的输入来增强模型的空间感知能力,并输出采样点相对于模型部件的内外状态。最后使用max pooling聚合解码器生成的隐式场,以得到模型的协同分割结果。结果实验结果表明,本文算法在Shape Net Part数据集上的m Io U(mean intersection-overunion)为62.1%,与目前已知的两类无监督3维点云模型分割方法相比,分别提高了22.5%和18.9%,分割性能得到了极大提升。与两种有监督方法相比,分别降低了19.3%和20.2%,但其在部件数较少的模型上可获得更优的分割效果。相比于使用交叉熵函数作为重构损失函...     

The shape variational autoencoder: A deep generative model of part‐segmented 3D objects

We introduce a generative model of part‐segmented 3D objects: the shape variational auto‐encoder (ShapeVAE). The ShapeVAE describes a joint distribution over the existence of object parts, the locations of a dense set of surface points, and over surface normals associated with these points. Our model makes use of a deep encoder‐decoder architecture that leverages the part‐decomposability of 3D objects to embed high‐dimensional shape representations and sample novel instances. Given an input collection of part‐segmented objects with dense point correspondences the ShapeVAE is capable of synthesizing novel, realistic shapes, and by performing conditional inference enables imputation of missing parts or surface normals. In addition, by generating both points and surface normals, our model allows for the use of powerful surface‐reconstruction methods for mesh synthesis. We provide a quantitative evaluation of the ShapeVAE on shape‐completion and test‐set log‐likelihood tasks and demonstrate that the model performs favourably against strong baselines. We demonstrate qualitatively that the ShapeVAE produces plausible shape samples, and that it captures a semantically meaningful shape‐embedding. In addition we show that the ShapeVAE facilitates mesh reconstruction by sampling consistent surface normals.     

Sketch‐to‐Design: Context‐Based Part Assembly

Designing 3D objects from scratch is difficult, especially when the user intent is fuzzy and lacks a clear target form. We facilitate design by providing reference and inspiration from existing model contexts. We rethink model design as navigating through different possible combinations of part assemblies based on a large collection of pre‐segmented 3D models. We propose an interactive sketch‐to‐design system, where the user sketches prominent features of parts to combine. The sketched strokes are analysed individually, and more importantly, in context with the other parts to generate relevant shape suggestions via adesign galleryinterface. As a modelling session progresses and more parts get selected, contextual cues become increasingly dominant, and the model quickly converges to a final form. As a key enabler, we use pre‐learned part‐based contextual information to allow the user to quickly explore different combinations of parts. Our experiments demonstrate the effectiveness of our approach for efficiently designing new variations from existing shape collections.     

Projective Feature Learning for 3D Shapes with Multi‐View Depth Images

Feature learning for 3D shapes is challenging due to the lack of natural paramterization for 3D surface models. We adopt the multi‐view depth image representation and propose Multi‐View Deep Extreme Learning Machine (MVD‐ELM) to achieve fast and quality projective feature learning for 3D shapes. In contrast to existing multi‐view learning approaches, our method ensures the feature maps learned for different views are mutually dependent via shared weights and in each layer, their unprojections together form a valid 3D reconstruction of the input 3D shape through using normalized convolution kernels. These lead to a more accurate 3D feature learning as shown by the encouraging results in several applications. Moreover, the 3D reconstruction property enables clear visualization of the learned features, which further demonstrates the meaningfulness of our feature learning.     

Robust Structure‐Based Shape Correspondence

We present a robust method to find region‐level correspondences between shapes, which are invariant to changes in geometry and applicable across multiple shape representations. We generate simplified shape graphs by jointly decomposing the shapes, and devise an adapted graph‐matching technique, from which we infer correspondences between shape regions. The simplified shape graphs are designed to primarily capture the overall structure of the shapes, without reflecting precise information about the geometry of each region, which enables us to find correspondences between shapes that might have significant geometric differences. Moreover, due to the special care we take to ensure the robustness of each part of our pipeline, our method can find correspondences between shapes with different representations, such as triangular meshes and point clouds. We demonstrate that the region‐wise matching that we obtain can be used to find correspondences between feature points, reveal the intrinsic self‐similarities of each shape and even construct point‐to‐point maps across shapes. Our method is both time and space efficient, leading to a pipeline that is significantly faster than comparable approaches. We demonstrate the performance of our approach through an extensive quantitative and qualitative evaluation on several benchmarks where we achieve comparable or superior performance to existing methods.     

A Revisit of Shape Editing Techniques:From the Geometric to the Neural Viewpoint

3D shape editing is widely used in a range of applications such as movie production,computer games and computer aided design.It is also a popular research topic in computer graphics and computer vision.In past decades,researchers have developed a series of editing methods to make the editing process faster,more robust,and more reliable.Traditionally,the deformed shape is determined by the optimal transformation and weights for an energy formulation.With increasing availability of 3D shapes on the Internet,data-driven methods were proposed to improve the editing results.More recently as the deep neural networks became popular,many deep learning based editing methods have been developed in this field,which are naturally data-driven.We mainly survey recent research studies from the geometric viewpoint to those emerging neural deformation techniques and categorize them into organic shape editing methods and man-made model editing methods.Both traditional methods and recent neural network based methods are reviewed.     

A review of solid shape modelling based on integrity verification

The potential benefits of using a canonical model for representing the shapes of solid objects has led to the design and implementation of a number of geometric modellers, with varying capabilities. This paper reviews the approaches taken in solid modelling by defining the well-formedness conditions which must be implemented in any modelling system. The methods for satisfying the well-formedness constraints in the various solid modelling methods are reviewed, using the concept of integrity constraint. The incorporation of integrity constraints, both explicitly and implicitly, into solid modellers are considered, with particular focus on boundary modellers. The use of integrity constraints for defining shape families and assembly families are also presented. The result is a unified view of solid shape modelling systems that enables their classification and extension into particular application areas.     

Markerless Multiview Motion Capture with 3D Shape Model Adaptation

In this paper, we address simultaneous markerless motion and shape capture from 3D input meshes of partial views onto a moving subject. We exploit a computer graphics model based on kinematic skinning as template tracking model. This template model consists of vertices, joints and skinning weights learned a priori from registered full‐body scans, representing true human shape and kinematics‐based shape deformations. Two data‐driven priors are used together with a set of constraints and cues for setting up sufficient correspondences. A Gaussian mixture model‐based pose prior of successive joint configurations is learned to soft‐constrain the attainable pose space to plausible human poses. To make the shape adaptation robust to outliers and non‐visible surface regions and to guide the shape adaptation towards realistically appearing human shapes, we use a mesh‐Laplacian‐based shape prior. Both priors are learned/extracted from the training set of the template model learning phase. The output is a model adapted to the captured subject with respect to shape and kinematic skeleton as well as the animation parameters to resemble the observed movements. With example applications, we demonstrate the benefit of such footage. Experimental evaluations on publicly available datasets show the achieved natural appearance and accuracy.     

Field functions for implicit surfaces

The use of 3D computer generated models is a rapidly growing part of the animation industry. But the established modelling techniques, using polygons or parametric patches, are not the best to define characters which can change their shape as they move. A newer method, using iso-surfaces in a scalar field, enables us to create models that can make the dynamic shape changes seen in hand animation. We call such modelsSoft Objects.From the user's point of view, a soft object is built from primitive key objects that blend to form a compound shape. In this paper, we examine some of the problems of choosing suitable keys and introduce some new field functions that increase the range of shapes available as keys.     

A Convolutional Decoder for Point Clouds using Adaptive Instance Normalization

Automatic synthesis of high quality 3D shapes is an ongoing and challenging area of research. While several data‐driven methods have been proposed that make use of neural networks to generate 3D shapes, none of them reach the level of quality that deep learning synthesis approaches for images provide. In this work we present a method for a convolutional point cloud decoder/generator that makes use of recent advances in the domain of image synthesis. Namely, we use Adaptive Instance Normalization and offer an intuition on why it can improve training. Furthermore, we propose extensions to the minimization of the commonly used Chamfer distance for auto‐encoding point clouds. In addition, we show that careful sampling is important both for the input geometry and in our point cloud generation process to improve results. The results are evaluated in an auto‐encoding setup to offer both qualitative and quantitative analysis. The proposed decoder is validated by an extensive ablation study and is able to outperform current state of the art results in a number of experiments. We show the applicability of our method in the fields of point cloud upsampling, single view reconstruction, and shape synthesis.