Computational image aesthetic evaluation is a computable human aesthetic perception and judgment realized by machines, which has a significant impact on a variety of applications such as image advanced search and promotional exhibition of painting arts. Various approaches have been proposed in copious literature trying to solve this challenging problem. However, there have been few attempts in reviewing works from different types of visual arts, due to their significant differences in visual features and aesthetic principles. In this survey, we present a comprehensive listing of the reviewed works on aesthetic assessment of photographs and paintings, mainly highlighting the contributions and innovations of the existing approaches. We firstly introduce aesthetic assessment benchmark datasets in different categories. Then, conventional aesthetic evaluation approaches based on handcrafted features are reviewed. Besides, we systematically evaluate recent deep learning techniques that are useful for developing robust models for aesthetic prediction tasks in scoring, distribution, attribute, and description. Moreover, the possibility of aesthetic-aware color enhancement, recomposition of photo images, and automatic generation of aesthetic-guided art paintings through computational approaches are summarized. Finally, challenges and potential future directions for this field are discussed. We hope that our survey could serve as a comprehensive reference source for future research on computational aesthetics in visual media.
Point cloud segmentation is an essential task in three-dimensional (3D) vision and intelligence. It is a critical step in understanding 3D scenes with a variety of applications. With the rapid development of 3D scanning devices, point cloud data have become increasingly available to researchers. Recent advances in deep learning are driving advances in point cloud segmentation research and applications. This paper presents a comprehensive review of recent progress in point cloud segmentation for understanding 3D indoor scenes. First, we present public point cloud datasets, which are the foundation for research in this area. Second, we briefly review previous segmentation methods based on geometry. Then, learning-based segmentation methods with multi-views and voxels are presented. Next, we provide an overview of learning-based point cloud segmentation, ranging from semantic segmentation to instance segmentation. Based on the annotation level, these methods are categorized into fully supervised and weakly supervised methods. Finally, we discuss open challenges and research directions in the future.
Assessing the aesthetic appeal of artworks has become an active research direction recently. However, previous works mainly focus on photographs and oil paintings, there have been few attempts in predicting aesthetics of Chinese ink paintings, due to their significant differences in visual features, semantic features, and aesthetic principles. Aiming at this problem, we propose a comprehensive framework, named Inkthetics, to quantify aesthetics of Chinese ink paintings based on deep learning. Firstly, an aesthetic assessment dataset is built for Chinese ink painting images. Secondly, a deep multi-view parallel convolutional neural network (DMVCNN) is designed by extracting global attribute images and multi-patches as inputs to jointly learn aesthetic features. Finally, we build a comprehensive aesthetic evaluation model by fusing the deeply-learned features with handcrafted features that rely on art expert knowledge. Experimental results show that our proposed deep network significantly outperforms existing methods on the dataset, and our proposed model can predict human aesthetic judgment with Pearson highly significant correlation of 0.843, which indicates an improvement up to 5.7% than the DMVCNN model when the handcrafted features are fused with activation from DMVCNN. Our work not only provides a deep-learning-based reference framework for computational aesthetic evaluation of Chinese paintings, but also explores to what extent can handcrafted features aid learning-based features in predicting human aesthetic perceptions.
Purpose: The purpose of this paper is to propose an interactive 2D–3D garment parametric pattern-making and linkage editing scheme that integrates clothing design, simulation and interaction to design 3D garments and 2D patterns. The proposed scheme has the potential to satisfy the individual needs of fashion industry, such as precise fit evaluation of the garment, interactive style editing with ease allowance and constrained contour lines in fashion design.Design/methodology/approach: The authors first construct a parametric pattern-making model for flat pattern design corresponding to the body dimensions. Then, the designing 2D patterns are stitched on a virtual 3D mannequin by performing a virtual try-on. If the customer is unsatisfied after the virtual try-on, the adjustable parameters (appearance parameters and fit parameters) can be adjusted using the 2D–3D linkage editing with hierarchical constrained contour lines, and the fit evaluation tool interactively provides the feedback.Findings: The authors observed that the usability and efficiency of the existing garment pattern-making method simplifies the garment pattern-making process. The authors utilize an interactive garment parametric flat pattern-making model to generate an individualized garment flat pattern that effectively adjust and realize the local editing of the garment pattern-making. The 2D–3D linkage editing is then employed, which alters the size and shape of garment pattern for a precise human model fit of the 3D garment using hierarchical constrained contour lines. Various instances have validated the effectiveness of the proposed scheme, which can increase the reusability of the existing garment styles and improve the efficiency of fashion design.Research limitations/implications: First, the authors do not consider the garment pattern-making design of sophisticated styles. Second, the authors do not directly consider complex garment shapes such as wrinkles, folds, multi-layer models and fabric physical properties.Originality/value: The authors propose a pattern adjustment scheme that uses the 3D virtual try-on technology to avoid repetitions of reality-based fit tests and garment sample making in the designing process of clothing products. The proposed scheme provides interactive selections of garment patterns and sizes and renders modification tools for 3D garment designing and 2D garment pattern-making. The authors present the 2D–3D interactive linkage editing scheme for a custom-fit garment pattern based on the hierarchical constraint contour lines. The spatial relationship among the human body, pattern pieces and 3D garment model is adequately expressed, and the final design result of the garment pattern is obtained by constraint solving. Meanwhile, the tightness tension of different parts of the 3D garment is analyzed, and the fit and comfort of the garment are quantitatively evaluated.
Cross-domain clothing retrieval is an active research topic because of its massive potential applications in fashion industry. Due to the large number of garment categories or styles, and different clothing appearances caused by different camera angles, different shooting conditions, different messy background environments, or different postures of the dressed human body, the retrieval accuracy of traditional consumer-to-shop scheme is always low. In this paper, based on the framework of deep convolution neural network, a novel cross-domain clothing retrieval method is proposed by using feature fusion and quadruplet loss function, which is named as ClothingNet. First, the pre-trained deep neural network Resnet-50 is adopted to extract feature map of clothing images. The extracted high-level image features can thus be merged with middle-level features, and the final feature representation of clothing images can be obtained by constraining the fusion feature values to a certain range in term of L2 norm. This fusion feature provides a comprehensive description of the differences between clothing images. For effectively training our ClothingNet, the cross-domain clothing images are organized in form of a quadruplet for calculating its loss function, and the network parameters can be optimized according to back propagation scheme via stochastic gradient descent of loss function. Our proposed method is validated on two public datasets for clothing retrieval, DARN and DeepFashion, showing that the top-50 retrieval accuracy is 35.67% and 53.52% respectively. Experimental results illustrate the effectiveness of our clothing retrieval method.