Author name cluster

Jun Wen

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

8 papers

2 author rows

EAAI Journal 2025 Journal Article

Reconstructing high-resolution flow fields from low-resolution experimental data based on multi-fidelity physics-informed neural network

Fan Zhang
Jiangli Chen
Zhenlin Xie
Jun Wen
Haibao Hu

High-resolution (HR) data are essential for fluid dynamic research, but obtaining HR experimental data is expensive currently. For particle image velocimetry (PIV) experiments, the most commonly used measurement method, low-resolution (LR) velocity fields are easily available, and HR data can be reconstructed from LR ones. Physics-informed neural network (PINN) is a framework that combines data-driven neural networks and physical laws, and we introduce a variant of PINN, the multi-fidelity PINN (MPINN), for the super-resolution reconstruction of LR data, which can decompose the data correlation into linear and nonlinear parts. Due to the lack of pressure data from PIV experiments, the physical equations in MPINN use the Navier–Stokes equations in the vortex-velocity form rather than the velocity–pressure form, and we call this network MPINN-V. By learning the distribution of HR data at low Reynolds numbers, the network can reconstruct the LR flow field at higher Reynolds numbers. First, model validation was conducted to investigate the effectiveness of selecting the vorticity–velocity formulation, as well as the influence of boundary conditions, network size, activation functions, data sparsity, and Reynolds number on the model’s reconstruction accuracy. Additionally, the super-resolution reconstruction capabilities of MPINN-V, PINN-V, and the traditional bicubic interpolation method were compared, with MPINN-V demonstrating higher accuracy. In addition, the network is robust to noise and can ignore 2. 5% Gaussian noise. Subsequently, we generalize the network to real experimental data, and the training dataset is still from numerical simulations. The results show that the network can achieve high-resolution reconstruction of experimental LR data, and at the same time, it also has the effect of noise reduction. All the above results show that MPINN can be well applied to flow field super-resolution reconstruction, and the vortex-velocity form NS equation can meet the needs of super-resolution reconstruction of experimental flow fields with only velocity data.

Details DOI

AAAI Conference 2021 Conference Paper

Context-Guided Adaptive Network for Efficient Human Pose Estimation

Lei Zhao
Jun Wen
Pengfei Wang
Nenggan Zheng

Although recent work has achieved great progress in human pose estimation (HPE), most methods show limitations in either inference speed or accuracy. In this paper, we propose a fast and accurate end-to-end HPE method, which is specifically designed to overcome the commonly encountered jitter box, defective box and ambiguous box problems of boxbased methods, e. g. Mask R-CNN. Concretely, 1) we propose the ROIGuider to aggregate box instance features from all feature levels under the guidance of global context instance information. Further, 2) the proposed Center Line Branch is equipped with a Dichotomy Extended Area algorithm to adaptively expand each instance box area, and Ambiguity Alleviation strategy to eliminate duplicated keypoints. Finally, 3) to achieve efficient multi-scale feature fusion and real-time inference, we design a novel Trapezoidal Network (TNet) backbone. Experimenting on the COCO dataset, our method achieves 68. 1 AP at 25. 4 fps, and outperforms Mask- RCNN by 8. 9 AP at a similar speed. The competitive performance on the HPE and person instance segmentation tasks over the state-of-the-art models show the promise of the proposed method. The source code will be made available at https: //github. com/zlcnup/CGANet.

PDF Details

AAAI Conference 2020 Conference Paper

Linear Context Transform Block

Dongsheng Ruan
Jun Wen
Nenggan Zheng
Min Zheng

Squeeze-and-Excitation (SE) block presents a channel attention mechanism for modeling global context via explicitly capturing dependencies across channels. However, we are still far from understanding how the SE block works. In this work, we ﬁrst revisit the SE block, and then present a detailed empirical study of the relationship between global context and attention distribution, based on which we propose a simple yet effective module, called Linear Context Transform (LCT) block. We divide all channels into different groups and normalize the globally aggregated context features within each channel group, reducing the disturbance from irrelevant channels. Through linear transform of the normalized context features, we model global context for each channel independently. The LCT block is extremely lightweight and easy to be plugged into different backbone models while with negligible parameters and computational burden increase. Extensive experiments show that the LCT block outperforms the SE block in image classiﬁcation task on the ImageNet and object detection/segmentation on the COCO dataset with different backbone models. Moreover, LCT yields consistent performance gains over existing state-of-the-art detection architectures, e. g. , 1. 5∼1. 7% APbbox and 1. 0%∼1. 2% APmask improvements on the COCO benchmark, irrespective of different baseline models of varied capacities. We hope our simple yet effective approach will shed some light on future research of attention-based models.

PDF Details

IJCAI Conference 2019 Conference Paper

Bayesian Uncertainty Matching for Unsupervised Domain Adaptation

Jun Wen
Nenggan Zheng
Junsong Yuan
Zhefeng Gong
Changyou Chen

Domain adaptation is an important technique to alleviate performance degradation caused by domain shift, e. g. , when training and test data come from different domains. Most existing deep adaptation methods focus on reducing domain shift by matching marginal feature distributions through deep transformations on the input features, due to the unavailability of target domain labels. We show that domain shift may still exist via label distribution shift at the classifier, thus deteriorating model performances. To alleviate this issue, we propose an approximate joint distribution matching scheme by exploiting prediction uncertainty. Specifically, we use a Bayesian neural network to quantify prediction uncertainty of a classifier. By imposing distribution matching on both features and labels (via uncertainty), label distribution mismatching in source and target data is effectively alleviated, encouraging the classifier to produce consistent predictions across domains. We also propose a few techniques to improve our method by adaptively reweighting domain adaptation loss to achieve nontrivial distribution matching and stable training. Comparisons with state of the art unsupervised domain adaptation methods on three popular benchmark datasets demonstrate the superiority of our approach, especially on the effectiveness of alleviating negative transfer.

PDF Details

AAAI Conference 2019 Conference Paper

Exploiting Local Feature Patterns for Unsupervised Domain Adaptation

Jun Wen
Risheng Liu
Nenggan Zheng
Qian Zheng
Zhefeng Gong
Junsong Yuan

Unsupervised domain adaptation methods aim to alleviate performance degradation caused by domain-shift by learning domain-invariant representations. Existing deep domain adaptation methods focus on holistic feature alignment by matching source and target holistic feature distributions, without considering local features and their multi-mode statistics. We show that the learned local feature patterns are more generic and transferable and a further local feature distribution matching enables fine-grained feature alignment. In this paper, we present a method for learning domain-invariant local feature patterns and jointly aligning holistic and local feature statistics. Comparisons to the state-of-the-art unsupervised domain adaptation methods on two popular benchmark datasets demonstrate the superiority of our approach and its effectiveness on alleviating negative transfer.

PDF Details

AAAI Conference 2018 Conference Paper

Unsupervised Representation Learning With Long-Term Dynamics for Skeleton Based Action Recognition

Nenggan Zheng
Jun Wen
Risheng Liu
Liangqu Long
Jianhua Dai
Zhefeng Gong

In recent years, skeleton based action recognition is becoming an increasingly attractive alternative to existing videobased approaches, beneﬁcial from its robust and comprehensive 3D information. In this paper, we explore an unsupervised representation learning approach for the ﬁrst time to capture the long-term global motion dynamics in skeleton sequences. We design a conditional skeleton inpainting architecture for learning a ﬁxed-dimensional representation, guided by additional adversarial training strategies. We quantitatively evaluate the effectiveness of our learning approach on three well-established action recognition datasets. Experimental results show that our learned representation is discriminative for classifying actions and can substantially reduce the sequence inpainting errors.

PDF Details

ICRA Conference 2015 Conference Paper

Automated micro-aspiration of mouse embryo limb bud tissue

Jun Wen
Jun Liu 0007
Kimberly Lau
Haijiao Liu
Sevan Hopyan
Yu Sun 0001

Mechanical force is an integral part of tissue morphogenesis and patterning. We have developed an automated micro-aspiration system to investigate how mouse limb bud tissue responds to extrinsic forces in order to understand whether tissue-generated forces can be a part of the mechanism causing oriented cell behaviors observed in mouse limb bud morphogenesis. The system is capable of performing automated micropipette tracking, tissue-tip contact detection, pressure control, and prolonged application of constant pressure. A three-dimensional tissue tracking algorithm is developed based on the processing of time-lapsed confocal Z-stack images. 3D visual feedback from confocal microscopy imaging, for the first time, is used to realize 3D visual servoing to control the micropipette position to compensate for tissue movement. This enables stable force application in a biologically relevant time scale (e. g. , 60 minutes) during which cell remodeling occurs. Experimental results demonstrate that micro-aspiration on mouse limb bud is capable of creating tension anisotropy which causes force-responsive cells to dynamically remodel through polarized cell division and rosette resolution.

Details

ICRA Conference 2015 Conference Paper

Automated robotic vitrification of embryos

Jun Liu 0007
Chaoyang Shi
Jun Wen
Derek Pyne
Haijiao Liu
Changhai Ru
Yu Sun 0001

This paper reports the first robotic system for vitrification of mammalian embryos. Vitrification is a technique for preserving oocytes and embryos in clinical IVF (in vitro fertilization). The procedure involves multiple steps of stringently timed pick-and-place operation for processing an oocyte/embryo in vitrification media. In IVF clinics, vitrification is conducted manually by highly skilled embryologists. Processing one oocyte/embryo occupies the embryologist 15–20 minutes, depending on protocols chosen to implement. Due to poor reproducibility and inconsistency across operators, success rates and survival rates also vary significantly. Through collaboration with IVF clinics, we are in process to realize robotic vitrification and aim ultimately to standardize clinical vitrification from manual operation to fully automated robotic operation. Our robotic system is embedded with two contact detection methods to determine the relative Z positions of the vitrification micropipette, embryo, and vitrification straw. A 3D tracking algorithm is developed for visually servoed embryo transfer and real-time monitoring of embryo volume changes during vitrification. Excess medium is automatically removed from around the vitrified embryo on the vitrification straw to achieve a high cooling rate. Tests on mouse embryos demonstrate that the system is capable of performing vitrification with a throughput at least three times that of manual operation and achieved a high survival rate (88. 9%) and development rate (93. 8%).

Details