Article
Robotics
Yan Wang, Hongwei Ma
Summary: We propose mVIL-Fusion, a three-level multisensor fusion system that achieves robust state estimation and globally consistent mapping in perceptually degraded environments. Our system uses LiDAR depth-assisted visual-inertial odometry (VIO) as the frontend, with synchronous prediction and distortion correction functions. It also applies a novel double-sliding-window-based optimization to enhance state estimation accuracy and robustness. Loop closures and pose-only factor graph smoothing are used in the backend to generate a global map. The system has been validated on public datasets and self-collected sequences.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2023)
Article
Chemistry, Analytical
Yinchu Wang, Haijiang Zhu
Summary: This paper proposes a lightweight network, CNNapsule, to solve the problem of weak adaptability to angle transformation in current monocular depth estimation algorithms. By integrating convolutional neural networks and matrix capsule features, as well as designing a specific loss function, the network improves adaptability and estimation accuracy.
Article
Environmental Sciences
Xiaofei Yang, Rencan Nie, Gucheng Zhang, Luping Chen, He Li
Summary: Pansharpening is a technology that fuses a low spatial resolution MS image with its associated high spatial full resolution PAN image. We propose a novel multistage Dense-Parallel attention fusion network (DPAFNet), which utilizes parallel attention residual dense block (PARDB) modules and multistage feature fusion to better focus on and exploit the intrinsic features and correlation between images, resulting in superior fusion results.
Article
Computer Science, Artificial Intelligence
Jingyu Chen, Xin Yang, Qizeng Jia, Chunyuan Liao
Summary: This study demonstrates that learning a convolutional neural network for depth estimation, along with an auxiliary optical flow network and epipolar geometry constraint, can greatly improve accuracy and speed. The architecture consists of tightly coupled encoder-decoder networks for optical flow and depth, with exchange blocks and an epipolar feature layer to enhance predictions. The method DENAO runs at high speed on a single GPU, outperforming state-of-the-art depth estimation and optical flow methods.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Article
Computer Science, Artificial Intelligence
Gao Huang, Zhuang Liu, Geoff Pleiss, Laurens van der Maaten, Kilian Q. Weinberger
Summary: Recent work has shown that adding shorter connections in convolutional networks can make the network deeper, more accurate, and more efficient in training. This paper introduces Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward manner. DenseNets alleviate the vanishing-gradient problem, encourage feature reuse, and improve parameter efficiency, leading to significant improvements in object recognition tasks.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Lei Han, Siyuan Gu, Dawei Zhong, Shuxue Quan, Lu Fang
Summary: This paper presents an RGBD-based globally-consistent dense 3D reconstruction approach that allows real-time geometric reconstruction and texture mapping, achieving realistic visualization with compact memory consumption and low computational complexity.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Multidisciplinary Sciences
Zhiyang Zhang, Shihua Zhang
Summary: This study presents mathematically equivalent forms of advanced deep learning models such as residual neural networks and dilated dense neural networks from the perspective of convolutional sparse coding. By considering factors such as initialization, dictionary design, and number of iterations, the researchers proposed novel multilayer models that have been proven effective through extensive numerical experiments and comparisons with competing methods. The study also provides a clear mathematical understanding of skip connections, dilated convolution, and dense connections in these models.
NATIONAL SCIENCE REVIEW
(2021)
Article
Computer Science, Artificial Intelligence
Guoquan Jiang, Rui Wu, Zhanqiang Huo, Cuijun Zhao, Junwei Luo
Summary: This study proposes a Lightweight Multi-Scale Adaptive Network (LigMSANet) to address the challenges of scale variation and real-time counting in highly congested scenes. The method breaks the scale limitation and adjusts the proportion of neurons with different receptive field sizes through a novel multi-scale adaptation module. By replacing standard convolution with depthwise separable convolution and using a tailored MobileNetV2, the model achieves improved performance with fewer parameters and runtimes.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Computer Science, Software Engineering
Jiazhao Zhang, Chenyang Zhu, Lintao Zheng, Kai Xu
Summary: The method proposed in this study tackles the challenges of fast-motion camera tracking using random optimization techniques, particularly Particle Filter Optimization (PFO). By updating a particle swarm template (PST), the method can efficiently drive thousands of particles to quickly and robustly locate and cover a good local optimum. The evaluation metric based on depth-model conformance allows the method to effectively track camera poses under fast motion, mitigating the effects of motion blur.
ACM TRANSACTIONS ON GRAPHICS
(2021)
Article
Engineering, Chemical
Xin Wu, Xiaoyan Liu, Jiaxu Duan
Summary: Accurately measuring the pellet size distribution during the pelletizing process is critical for blast furnace efficiency. A method based on a convolutional neural network is proposed to measure the pellet size distribution, which outperforms other methods and meets the requirements of online measurement.
Article
Engineering, Multidisciplinary
Yan Su, Lei Yu
Summary: This paper presents a CNN-based SLAM reconstruction algorithm that optimizes feature point extraction and pose estimation, resulting in a complete and smooth spatial model.
MEASUREMENT SCIENCE AND TECHNOLOGY
(2022)
Article
Computer Science, Information Systems
Fasih Ud Din Farrukh, Weiyi Zhang, Chun Zhang, Zhihua Wang, Hanjun Jiang
Summary: A hardware architecture for a feature point extraction method based on deep neural networks for simultaneous localization and mapping in robotic applications is proposed. Key techniques are deployed to improve hardware and power efficiency, achieving significant progress in reducing memory overhead and on-chip storage.
Article
Engineering, Marine
Antoni Burguera, Francisco Bonin-Font, Eric Guerrero Font, Antoni Martorell Torres
Summary: This paper presents a new approach to visual Graph-SLAM for underwater robots, focusing on designing a robust VLD algorithm to reduce false loops and improve computational efficiency. The method operates in three steps, utilizing a Neural Network for fast selection of image pairs, a robust image matcher for loop confirmation or rejection, and geometric consistency verification for accepted loops. Results show the validity and suitability of the approach for further field campaigns.
JOURNAL OF MARINE SCIENCE AND ENGINEERING
(2022)
Article
Computer Science, Artificial Intelligence
Yongzhi Long, Haitao Jia, Yida Zhong, Yadong Jiang, Yuming Jia
Summary: This study introduces a novel unsupervised network called RXDNFuse for infrared/visible light fusion task, utilizing a combination of ResNeXt and DenseNet structures. The method automatically estimates information preservation levels and incorporates loss function strategies for network parameter training, improving quality of detailed information. The results demonstrate effective preservation of textural details and thermal radiation information, aligning well with human visual perception system.
INFORMATION FUSION
(2021)
Article
Multidisciplinary Sciences
Yasuhide Hirohata, Maina Sogabe, Tetsuro Miyazaki, Toshihiro Kawase, Kenji Kawashima
Summary: This paper presents a method to accurately estimate depth from monocular laparoscopic images in dynamic surgical environments. The lack of reliable ground truth and the presence of noise elements like bleeding and smoke make this task complex. The proposed model learning framework uses a laparoscopic surgery video dataset for training and employs a unique loss function to ensure robust learning and precise depth estimation.
SCIENTIFIC REPORTS
(2023)
Article
Nanoscience & Nanotechnology
Karl Berggren, Qiangfei Xia, Konstantin K. Likharev, Dmitri B. Strukov, Hao Jiang, Thomas Mikolajick, Damien Querlioz, Martin Salinga, John R. Erickson, Shuang Pi, Feng Xiong, Peng Lin, Can Li, Yu Chen, Shisheng Xiong, Brian D. Hoskins, Matthew W. Daniels, Advait Madhavan, James A. Liddle, Jabez J. McClelland, Yuchao Yang, Jennifer Rupp, Stephen S. Nonnenmann, Kwang-Ting Cheng, Nanbo Gong, Miguel Angel Lastras-Montano, A. Alec Talin, Alberto Salleo, Bhavin J. Shastri, Thomas Ferreira de Lima, Paul Prucnal, Alexander N. Tait, Yichen Shen, Huaiyu Meng, Charles Roques-Carmes, Zengguang Cheng, Harish Bhaskaran, Deep Jariwala, Han Wang, Jeffrey M. Shainline, Kenneth Segall, J. Joshua Yang, Kaushik Roy, Suman Datta, Arijit Raychowdhury
Summary: Recent progress in artificial intelligence is primarily attributed to the rapid development of machine learning, but the performance and energy efficiency of hardware systems set fundamental limits on machine learning capabilities. Data-centric computing requires a revolution in hardware systems, with new hardware platforms offering hope for future computing with improved throughput and energy efficiency. However, challenges such as materials selection, device optimization, circuit fabrication, and system integration must be addressed in building such systems.
Article
Computer Science, Information Systems
Zengqiang Yan, Jeffry Wicaksana, Zhiwei Wang, Xin Yang, Kwang-Ting Cheng
Summary: The paper introduces a variation-aware federated learning (VAFL) framework to address the cross-client variation problem in medical image data by minimizing variations among clients while preserving privacy, used for automated classification of clinically significant prostate cancer.
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS
(2021)
Review
Materials Science, Multidisciplinary
Yvan Bonnassieux, Christoph J. Brabec, Yong Cao, Tricia Breen Carmichael, Michael L. Chabinyc, Kwang-Ting Cheng, Gyoujin Cho, Anjung Chung, Corie L. Cobb, Andreas Distler, Hans-Joachim Egelhaaf, Gerd Grau, Xiaojun Guo, Ghazaleh Haghiashtiani, Tsung-Ching Huang, Muhammad M. Hussain, Benjamin Iniguez, Taik-Min Lee, Ling Li, Yuguang Ma, Dongge Ma, Michael C. McAlpine, Tse Nga Ng, Ronald osterbacka, Shrayesh N. Patel, Junbiao Peng, Huisheng Peng, Jonathan Rivnay, Leilai Shao, Daniel Steingart, Robert A. Street, Vivek Subramanian, Luisa Torsi, Yunyun Wu
Summary: This roadmap presents perspectives and visions from leading researchers in the fields of flexible and printable electronics, covering device technologies, fabrication techniques, and design and modeling approaches essential for future development of new applications leveraging flexible electronics (FE). It aims to serve as a resource on the current status and future challenges, highlighting the broad opportunities made available by FE technologies.
FLEXIBLE AND PRINTED ELECTRONICS
(2021)
Article
Computer Science, Hardware & Architecture
Dawen Xu, Meng He, Cheng Liu, Ying Wang, Long Cheng, Huawei Li, Xiaowei Li, Kwang-Ting Cheng
Summary: AIoT processors fabricated with newer technology nodes are susceptible to rising soft errors, especially in deep learning accelerators. To address this issue, a remote retraining framework and an optimized partial triple modular redundancy strategy are proposed. The experiments show that this approach allows for tradeoffs between model accuracy and performance penalty, while a data transmission optimization method reduces retraining time significantly.
IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS
(2021)
Article
Computer Science, Interdisciplinary Applications
Weihang Dai, Xiaomeng Li, Wan Hang Keith Chiu, Michael D. Kuo, Kwang-Ting Cheng
Summary: This paper proposes AdaCon, a contrastive learning framework for deep image regression, which incorporates a novel adaptive-margin contrastive loss and a regression prediction branch for feature learning. By considering label distance relationships in feature representations, AdaCon achieves better performance in downstream regression tasks. Experimental results on two medical image regression tasks demonstrate the effectiveness of AdaCon, with relative improvements of 3.3% and 5.9% in MAE compared to state-of-the-art methods for BMD estimation and LVEF prediction, respectively.
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2022)
Article
Computer Science, Artificial Intelligence
Yuanjie Dang, Chong Huang, Peng Chen, Ronghua Liang, Xin Yang, Kwang-Ting Cheng
Summary: This study proposes an integrated aerial filming system that autonomously captures cinematic shots of action scenes by imitating demonstrations. The system utilizes the deep deterministic policy gradient to build a model and designs a spatial attention network to selectively focus on discriminative joints of the skeleton. Experimental results demonstrate that our method successfully mimics viewpoint selection strategy and captures more accurate viewpoints compared to existing techniques.
IEEE TRANSACTIONS ON COGNITIVE AND DEVELOPMENTAL SYSTEMS
(2022)
Article
Computer Science, Hardware & Architecture
Cheng Liu, Cheng Chu, Dawen Xu, Ying Wang, Qianlong Wang, Huawei Li, Xiaowei Li, Kwang-Ting Cheng
Summary: This paper proposes a hybrid computing architecture for fault-tolerant DLAs, which shows significantly higher reliability, scalability, and performance with less chip area penalty compared to conventional redundancy approaches. By taking advantage of flexible recomputing, it can also be used to scan the entire 2-D computing array and effectively detect faulty PEs at runtime.
IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS
(2022)
Article
Computer Science, Information Systems
Jeffry Wicaksana, Zengqiang Yan, Xin Yang, Yang Liu, Lixin Fan, Kwang-Ting Cheng
Summary: The performance of deep networks for medical image analysis is often limited by the scarcity of medical data and privacy concerns. To address this issue, we propose CusFL, a customized federated learning approach that enables each client to train a client-specific model based on a federated global model. By using a federated feature extractor for guidance, CusFL allows clients to selectively learn useful knowledge from the federated model and improve their personalized models.
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Arnav Chavan, Zhiqiang Shen, Zhuang Liu, Zechun Liu, Kwang-Ting Cheng, Eric Xing
Summary: This paper explores the feasibility of finding an optimal sub-model from a vision transformer and introduces a pure vision transformer slimming (ViT-Slim) framework. The proposed method achieves high compression rates and accuracy improvements on various vision transformers through an end-to-end searching process across multiple dimensions.
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022)
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Yi Lin, Zeyu Wang, Kwang-Ting Cheng, Hao Chen
Summary: This paper proposes a realistic data augmentation method called InsMix for nuclei segmentation. The method utilizes morphology constraints and background perturbation to enhance the images, enabling rich information acquisition about the nuclei while preserving their morphology characteristics. Experimental results demonstrate the superior performance of the proposed method on two datasets.
MEDICAL IMAGE COMPUTING AND COMPUTER ASSISTED INTERVENTION, MICCAI 2022, PT II
(2022)
Proceedings Paper
Neuroimaging
Yu Cai, Hao Chen, Xin Yang, Yu Zhou, Kwang-Ting Cheng
Summary: This paper proposes a new method for anomaly detection using both known normal images and unlabeled images, achieving significant improvements on three CXR datasets.
MEDICAL IMAGE COMPUTING AND COMPUTER ASSISTED INTERVENTION, MICCAI 2022, PT III
(2022)
Article
Computer Science, Artificial Intelligence
Chong Huang, Yuanjie Dang, Peng Chen, Xin Yang, Kwang-Ting Cheng
Summary: Imitation learning is applied to autonomous camera systems, but current methods require a large number of training videos with similar styles and struggle to generalize to different styles. To address this, a framework called one-shot imitation filming is proposed, which can imitate a filming style without style-specific model training using two key techniques: filming style feature extraction and camera motion prediction.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Junlin Xian, Zhiwei Wang, Kwang-Ting Cheng, Xin Yang
Summary: A holistic understanding of dual-view transformation is important for computer-aided diagnosis of breast lesions, and densifying sparse supervision by synthesizing lesions across two views can lead to superior performance in cross-view lesion matching.
MEDICAL IMAGE COMPUTING AND COMPUTER ASSISTED INTERVENTION - MICCAI 2021, PT V
(2021)
Article
Computer Science, Artificial Intelligence
Zechun Liu, Xiangyu Zhang, Zhiqiang Shen, Yichen Wei, Kwang-Ting Cheng, Jian Sun
Summary: This study presents a joint multi-dimension pruning method, effectively pruning a network on three crucial aspects simultaneously. By defining the pruning vector, constructing a mapping from the vector to the pruned network structure, and optimizing the vector through numerical gradient optimization, the method collaboratively optimizes across dimensions and achieves better performance.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2021)
Article
Engineering, Civil
Xin Yang, Jingyu Chen, Yuanjie Dang, Hongcheng Luo, Yuesheng Tang, Chunyuan Liao, Peng Chen, Kwang-Ting Cheng
Summary: This paper presents a real-time onboard approach for monocular depth prediction and obstacle avoidance with a lightweight probabilistic CNN, which efficiently predicts depth and confidence, generates traversable waypoints, and produces control inputs for drones. Experimental results demonstrate that the method runs faster than state-of-the-art approaches and achieves better depth estimation accuracy, showing superiority in obstacle avoidance in simulated and real environments.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2021)