Article
Computer Science, Software Engineering
Sebastian Winberg, Gaspard Zoss, Prashanth Chandran, Paulo Gotardo, Derek Bradley
Summary: Facial hair is often overlooked in facial capture, and existing methods have issues in tracking 3D facial hair and underlying skin. In this paper, we propose a multiview reconstruction pipeline that tracks facial hair and skin for entire performances. Our method does not require dense camera arrays and can be applied to different facial hairstyles and lengths.
ACM TRANSACTIONS ON GRAPHICS
(2022)
Article
Computer Science, Information Systems
Do-Yeon Hwang, Seok-Hwan Choi, Jinmyeong Shin, Moonkyu Kim, Yoon-Ho Choi
Summary: This paper proposed a new deep learning-based image translation method for predicting and generating images before and after hair transplant surgery. By using a novel ROI image translation method based on generative adversarial networks (GAN), the model is able to convert only the region of interest (ROI) while retaining non-ROI regions. Experimental results demonstrated the superiority of this method in terms of SSIM, IoU, and FID metrics, and the ensemble method showed consistent performance in enhancing ROI detection and generating better predictive images.
Article
Computer Science, Software Engineering
Chufeng Xiao, Deng Yu, Xiaoguang Han, Youyi Zheng, Hongbo Fu
Summary: This paper presents a method for generating realistic hair images directly from hair sketches, incorporating a two-stage framework, self-attention modules, and sketch completion strategies to efficiently capture complex hair structures and appearance, reducing user workload.
ACM TRANSACTIONS ON GRAPHICS
(2021)
Article
Mathematics, Interdisciplinary Applications
Luo Dai
Summary: This article proposes a multi-feature fusion error action expression method to improve the effectiveness of human body error action recognition. The method, based on silhouette and optical flow information, avoids laborious preprocessing operations and achieves a higher recognition rate.
Article
Automation & Control Systems
Hongwei Guan, Lingjian Ye, Yurun Wang, Feifan Shen, Yuchen He
Summary: A dynamic model for the absorber in the MEA-based CO2 capture process is developed and validated using steady-state experimental data. Sensitivity analysis and a parameter estimation method are proposed to improve the model accuracy. Dynamic simulations investigate the effects of lean MEA solution feed rate and flue inlet conditions, providing new insights.
INTELLIGENT AUTOMATION AND SOFT COMPUTING
(2023)
Article
Engineering, Environmental
Federico Atzori, Francesco Barzagli, Alberto Varone, Giacomo Cao, Alessandro Concas
Summary: It is well known that CO2 capture and re-use is one of the main challenges to tackle global warming. Aqueous ammonia solutions have shown promising potential as sorbents for CO2 capture, offering higher absorption capacity, lower energy requirements, and greater resistance to degradation compared to conventional alkanolamines. This study investigates the chemical absorption of CO2 in aqueous NH3 solutions and proposes a mathematical model to quantify the capture efficiency dynamics and sorbent composition. The results validate the model's effectiveness in designing and optimizing the capture process, considering the effects of temperature and sorbent concentration.
CHEMICAL ENGINEERING JOURNAL
(2023)
Article
Computer Science, Software Engineering
Yuefan Shen, Changgeng Zhang, Hongbo Fu, Kun Zhou, Youyi Zheng
Summary: DeepSketchHair is a deep learning-based tool for modeling 3D hair from 2D sketches. The system utilizes three carefully designed neural networks to convert, map, and update input sketches, generating 3D hair models that match the input sketches.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2021)
Article
Engineering, Aerospace
Guobin Zhang, Qingbin Zhang, Zhiwei Feng, Qingquan Chen, Tao Yang
Summary: Flexible capture has high applicability in the removal of space debris. A novel adhesive capture mechanism is proposed in this paper, with detailed multibody dynamic modeling method and investigations of the adhesive capture process. The study discusses the impact of system parameters on the capture effectiveness and provides key parameters for selecting adhesive materials.
ADVANCES IN SPACE RESEARCH
(2021)
Article
Energy & Fuels
Simion Dragan, Hannelore Lisei, Flavia-Maria Ilea, Alexandru-Constantin Bozonc, Ana-Maria Cormos
Summary: In this study, a comprehensive mathematical model is presented to address climate change through the process of carbon dioxide capture using ammonia aqueous solutions. The model is validated and shows strong correlation with experimental results. Scaling up the model to simulate industrial-scale ammonia-based absorption process reveals its potential utility for future applications, including process optimization and control techniques for mitigating drawbacks. The model's ability to effectively replicate system behavior under various conditions highlights its importance.
Article
Medicine, General & Internal
Lintong Zhang, Qiaoyue Man, Young Im Cho
Summary: With the development of medicine and deep learning technologies, judging the degree of hair damage has become more important. There are currently three methods available, but all have certain limitations and inconveniences, leading to the proposal of a method combining scanning electron microscope and deep learning for improved accuracy.
Article
Computer Science, Artificial Intelligence
Zhen Li, Xinjiang Ye, Huawei Liang
Summary: This paper builds a sports video analysis system based on image recognition technology, which comprehensively analyzes dynamic images using various techniques to improve the data fusion effect, as well as designs a dynamic image analysis method to enhance the quality of the fused dynamic images.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Information Systems
Qiang Fu, Hanxiang Fu, Yuezhou Wu
Summary: This study proposes a network called MCDFN, which is based on mask and cross-dynamic fusion. By adaptingively preserving salient features of infrared images and texture details of visible images through an end-to-end fusion process, this network achieves good results on public datasets.
Article
Chemistry, Multidisciplinary
Feitong Wu, Yanping Du, Sijia Lv, Changying Zhao, Xiang Yang
Summary: Solar-assisted photocatalysis is considered an ideal option for a sustainable future of energy and environment due to its merits in carbon circulation and hydrocarbon production. This study investigates the micro-mechanism of CO2 adsorption and conversion in an Au-TiO2 photocatalytic system using density functional theory model, and explores the effects of water molecule proportion and temperature increase on the photocatalytic process. The results show that under certain experimental conditions, CO2 adsorption with HCOO center dot formation can occur without the need of activation energy.
Article
Computer Science, Artificial Intelligence
Yaowei Li, Jinshan Pan, Ye Luo, Jianwei Lu
Summary: This paper proposes an exemplar-based method to solve the problem of dynamic scene deblurring, using siamese encoder network and shallow encoder network to extract features and develop a rank module to explore useful features, achieving significant improvements in both quantitative and qualitative evaluations.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Information Systems
Sukesh Adiga, Jose Dolz, Herve Lombaert
Summary: This paper proposes a dynamic subspace learner that can dynamically utilize multiple learners to improve similarity learning in medical image analysis. The integration of an attention module enhances the visual interpretability of the method. The method achieves competitive results and outperforms other methods in image clustering, retrieval, and segmentation applications.
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS
(2022)
Article
Computer Science, Software Engineering
Wonjong Jang, Gwangjin Ju, Yucheol Jung, Jiaolong Yang, Xin Tong, Seungyong Lee
Summary: Our framework, StyleCariGAN, utilizes Shape and style manipulation with StyleGAN to automatically generate realistic and detailed caricatures with optional controls. Experimental results show that StyleCariGAN produces more realistic and detailed caricatures compared to current state-of-the-art methods. Furthermore, StyleCariGAN also supports other StyleGAN-based image manipulations, such as facial expression control.
ACM TRANSACTIONS ON GRAPHICS
(2021)
Article
Computer Science, Software Engineering
Yizhong Zhang, Jiaolong Yang, Zhen Liu, Ruicheng Wang, Guojun Chen, Xin Tong, Baining Guo
Summary: The VirtualCube system is a 3D video conference system that utilizes virtual cubicles and advanced rendering techniques to enable realistic interactions and eye contact with remote participants.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2022)
Editorial Material
Computer Science, Software Engineering
Ming C. Lin, Xin Tong, Wenping Wang
COMPUTATIONAL VISUAL MEDIA
(2022)
Article
Computer Science, Software Engineering
Inseung Hwang, Daniel S. Jeon, Adolfo Munoz, Diego Gutierrez, Xin Tong, Min H. Kim
Summary: Ellipsometry techniques are used to measure the polarization information of materials, but traditional methods are time-consuming and require cumbersome devices. This paper presents a sparse ellipsometry method that can capture both polarimetric reflectance information and the 3D shape of objects using a portable device. The results are in strong agreement with a ground-truth dataset of polarimetric BRDFs of real-world objects.
ACM TRANSACTIONS ON GRAPHICS
(2022)
Article
Computer Science, Software Engineering
Haoxiang Guo, Shilin Liu, Hao Pan, Yang Liu, Xin Tong, Baining Guo
Summary: This study views the reconstruction of CAD models as the detection of geometric primitives and their correspondence, and proposes a novel neural network framework for more complete and regularized reconstructions. By solving a global optimization and applying geometric refinements, it achieves more accurate and complete CAD B-Rep models.
ACM TRANSACTIONS ON GRAPHICS
(2022)
Article
Computer Science, Software Engineering
Peng-Shuai Wang, Yang Liu, Xin Tong
Summary: This paper presents an adaptive deep representation of volumetric fields of 3D shapes and an efficient approach to learn this representation for high-quality 3D shape reconstruction and auto-encoding. The method encodes the volumetric field with an adaptive feature volume organized by an octree and applies a compact multilayer perceptron network for mapping the features to the field value. The approach effectively encodes shape details, enables fast 3D shape reconstruction, and exhibits good generality.
ACM TRANSACTIONS ON GRAPHICS
(2022)
Article
Computer Science, Software Engineering
X. Zheng, Y. Liu, P. Wang, X. Tong
Summary: We propose a StyleGAN2-based deep learning approach, SDF-StyleGAN, for 3D shape generation. By extending StyleGAN2 to 3D generation and utilizing the implicit signed distance function as the shape representation, we introduce global and local shape discriminators to improve the geometry and visual quality of the shapes. We use shading-image-based Frechet inception distance scores to evaluate the visual quality and shape distribution of the generated shapes.
COMPUTER GRAPHICS FORUM
(2022)
Editorial Material
Computer Science, Information Systems
Xin Tong
FRONTIERS OF INFORMATION TECHNOLOGY & ELECTRONIC ENGINEERING
(2022)
Article
Computer Science, Artificial Intelligence
Xiaobin Hu, Wenqi Ren, Jiaolong Yang, Xiaochun Cao, David Wipf, Bjoern Menze, Xin Tong, Hongbin Zha
Summary: This paper proposes an improved face restoration method by embedding the network with 3D morphable priors, which enhances the performance of facial restoration tasks. Experimental results demonstrate superior performance of this method in face super-resolution and deblurring.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Software Engineering
Chun-Yu Sun, Yu-Qi Yang, Hao-Xiang Guo, Peng-Shuai Wang, Xin Tong, Yang Liu, Heung-Yeung Shum
Summary: The lack of fine-grained 3D shape segmentation data is a major challenge for developing learning-based 3D segmentation techniques. In this study, we propose an effective semi-supervised method that learns 3D segmentations using a combination of labeled and unlabeled data. Our approach incorporates a novel multilevel consistency loss to ensure consistent network predictions across different levels of perturbed 3D shapes, and a part substitution scheme to augment labeled data for improved training. Extensive validation on different tasks demonstrates the superior performance of our method compared to existing semi-supervised and unsupervised pre-training approaches.
COMPUTATIONAL VISUAL MEDIA
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Yu Deng, Jiaolong Yang, Jianfeng Xiang, Xin Tong
Summary: The study aims to generate 3D-consistent images with controllable camera poses through 3D-aware image generative modeling. A novel approach is proposed to regulate point sampling and radiance field learning on 2D manifolds, addressing the limitations in handling fine details and stable training in existing generators.
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)
(2022)
Article
Computer Science, Artificial Intelligence
Jiapeng Tang, Xiaoguang Han, Mingkui Tan, Xin Tong, Kui Jia
Summary: This paper focuses on the challenging task of learning 3D object surface reconstructions from RGB images and proposes a method that learns and uses a topology-preserved skeletal shape representation to assist the surface reconstruction. Through experiments, the proposed method is shown to be effective and outperforms existing methods in the surface reconstruction task.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Haofei Xu, Jiaolong Yang, Jianfei Cai, Juyong Zhang, Xin Tong
Summary: This paper proposes a new method for high-resolution optical flow estimation inspired by Transformers, using 1D attention and correlation operations to achieve 2D correspondence modeling effect and significantly reduce computational complexity. Experimental results demonstrate the effectiveness and superiority of the proposed method in handling high-resolution images.
2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021)
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Ming-Jia Yang, Yu-Xiao Guo, Bin Zhou, Xin Tong
Summary: The method presented in this study utilizes a generative model trained on semantic-segmented depth images to automatically generate 3D indoor scenes, modeling each scene as a 3D semantic volume and learning from 2.5D partial observations. Compared to existing methods, it reduces modeling and acquisition workload, producing improved object shapes and layouts.
2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021)
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Jongyoo Kim, Jiaolong Yang, Xin Tong
Summary: This study introduces a new method for completing invisible textures in single face images without using any complete textures, achieved through unsupervised learning using a large corpus of face images. The proposed DSD-GAN method utilizes two discriminators in UV map space and image space to learn both facial structures and texture details in a complementary manner, demonstrating the importance of their combination for high-fidelity results. Despite never seeing complete facial appearances, the network is able to generate compelling full textures from single images.
2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021)
(2021)