@article{xue_wu_xia_zhang_2022, title={Learning Local-Global Contextual Adaptation for Multi-Person Pose Estimation}, ISSN={["1063-6919"]}, DOI={10.1109/CVPR52688.2022.01272}, abstractNote={This paper studies the problem of multi-person pose estimation in a bottom-up fashion. With a new and strong observation that the localization issue of the center-offset formulation can be remedied in a local-window search scheme in an ideal situation, we propose a multi-person pose estimation approach, dubbed as LOGO-CAP, by learning the LOcal-GlObal Contextual Adaptation for human Pose. Specifically, our approach learns the keypoint attraction maps (KAMs) from the local keypoints expansion maps (KEMs) in small local windows in the first step, which are subsequently treated as dynamic convolutional kernels on the keypoints-focused global heatmaps for contextual adaptation, achieving accurate multi-person pose estimation. Our method is end-to-end trainable with near real-time inference speed in a single forward pass, obtaining state-of-the-art performance on the COCO keypoint benchmark for bottom-up human pose estimation. With the COCO trained model, our method also outperforms prior arts by a large margin on the challenging OCHuman dataset.}, journal={2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)}, author={Xue, Nan and Wu, Tianfu and Xia, Gui-Song and Zhang, Liangpei}, year={2022}, pages={13055–13064} }
@article{xue_wu_bai_wang_xia_zhang_torr_2020, title={Holistically-Attracted Wireframe Parsing}, ISSN={["1063-6919"]}, DOI={10.1109/CVPR42600.2020.00286}, abstractNote={This paper presents a fast and parsimonious parsing method to accurately and robustly detect a vectorized wireframe in an input image with a single forward pass. The proposed method is end-to-end trainable, consisting of three components: (i) line segment and junction proposal generation, (ii) line segment and junction matching, and (iii) line segment and junction verification. For computing line segment proposals, a novel exact dual representation is proposed which exploits a parsimonious geometric reparameterization for line segments and forms a holistic 4-dimensional attraction field map for an input image. Junctions can be treated as the “basins” in the attraction field. The proposed method is thus called Holistically-Attracted Wireframe Parser (HAWP). In experiments, the proposed method is tested on two benchmarks, the Wireframe dataset [14] and the YorkUrban dataset [8]. On both benchmarks, it obtains state-of-the-art performance in terms of accuracy and efficiency. For example, on the Wireframe dataset, compared to the previous state-of-the-art method L-CNN [36], it improves the challenging mean structural average precision (msAP) by a large margin (2.8% absolute improvements), and achieves 29.5 FPS on a single GPU (89% relative improvement). A systematic ablation study is performed to further justify the proposed method.}, journal={2020 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)}, author={Xue, Nan and Wu, Tianfu and Bai, Song and Wang, Fudong and Xia, Gui-Song and Zhang, Liangpei and Torr, Philip H. S.}, year={2020}, pages={2785–2794} }
@article{xue_bai_wang_xia_wu_zhang_2019, title={Learning Attraction Field Representation for Robust Line Segment Detection}, ISSN={["1063-6919"]}, DOI={10.1109/CVPR.2019.00169}, abstractNote={This paper presents a region-partition based attraction field dual representation for line segment maps, and thus poses the problem of line segment detection (LSD) as the region coloring problem. The latter is then addressed by learning deep convolutional neural networks (ConvNets) for accuracy, robustness and efficiency. For a 2D line segment map, our dual representation consists of three components: (i) A region-partition map in which every pixel is assigned to one and only one line segment; (ii) An attraction field map in which every pixel in a partition region is encoded by its 2D projection vector w.r.t. the associated line segment; and (iii) A squeeze module which squashes the attraction field to a line segment map that almost perfectly recovers the input one. By leveraging the duality, we learn ConvNets to compute the attraction field maps for raw in-put images, followed by the squeeze module for LSD, in an end-to-end manner. Our method rigorously addresses several challenges in LSD such as local ambiguity and class imbalance. Our method also harnesses the best practices developed in ConvNets based semantic segmentation methods such as the encoder-decoder architecture and the a-trous convolution. In experiments, our method is tested on the WireFrame dataset and the YorkUrban dataset with state-of-the-art performance obtained. Especially, we advance the performance by 4.5 percents on the WireFramedataset. Our method is also fast with 6.6∼10.4 FPS, outperforming most of existing line segment detectors.}, journal={2019 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2019)}, author={Xue, Nan and Bai, Song and Wang, Fudong and Xia, Gui-Song and Wu, Tianfu and Zhang, Liangpei}, year={2019}, pages={1595–1603} }
@inproceedings{xue_chakrabortty_2016, title={H-2-clustering of closed-loop consensus networks under a class of LQR design}, DOI={10.1109/acc.2016.7526799}, abstractNote={Given any positive integer r, our objective is to develop a strategy for grouping the states of a n-node network into r ≤ n distinct non-overlapping groups. The criterion for this partitioning is defined as follows. First, a LQR controller is defined for the original n-node network. Then, a r-dimensional reduced-order network is created by imposing a projection matrix P on the n-node open-loop network, and a reduced-order r-dimensional LQR controller is constructed. The resulting controller is, thereafter, projected back to its original coordinates, and implemented in the n-node network. The problem, therefore, is to find a grouping strategy or P that will minimize the difference between the closed-loop transfer matrix of the original network with the full-order controller and that with the projected controller, in the sense of H2 norm. We derive an upper bound on this difference in terms of P, and, thereby propose a design for P using K-means that tightens the bound while guaranteeing numerical feasibility.}, booktitle={2016 american control conference (acc)}, author={Xue, N. and Chakrabortty, Aranya}, year={2016}, pages={7141–7146} }
@inproceedings{xue_chakrabortty_2016, title={H-2-clustering of closed-loop consensus networks under generalized LQR designs}, DOI={10.1109/cdc.2016.7799051}, abstractNote={In this paper we present a Linear Quadratic Regulator (LQR) control design for large-scale consensus networks. When such networks have tens of thousands of nodes spread over a wide geographical span, the design and implementation of conventional LQR controllers become very challenging. Consider an n-node consensus network with both node and edge weights. Given any positive integer r, our objective is to develop a strategy for grouping the states of this network into r distinct non-overlapping groups. The criterion for this partitioning is defined as follows. First, an LQR state-feedback controller is defined over the n-node network for any given Q ≥ 0. Then, an r-dimensional reduced-order network is created by imposing a projection matrix P on the open-loop network, and a reduced-order r-dimensional LQR controller is constructed. The resulting controller is, thereafter, projected back to its original coordinates, and implemented in the n-node network. The problem, therefore, is to find a grouping strategy or P that will minimize the difference between the closed-loop transfer matrix of the original network with the full-order controller and that with the projected controller in the sense of ℋ2 norm. We derive an upper bound on this difference in terms of P, and, thereby propose a design for P using weighted k-means that tightens the bound. The weighting of k-means arises due to the node weights in the network, and the resulting asymmetry in its Laplacian matrix.}, booktitle={2016 ieee 55th conference on decision and control (cdc)}, author={Xue, N. and Chakrabortty, Aranya}, year={2016}, pages={5116–5121} }
@article{xue_chakrabortty_2016, title={Parallel Identification of Power System Dynamic Models Under Scheduling Constraints}, volume={31}, ISSN={["1558-0679"]}, DOI={10.1109/tpwrs.2015.2504453}, abstractNote={In this paper we present two sets of parallel algorithms for identifying real-time, small-signal dynamic models of power systems using multiple sources of Synchrophasor data. The first problem is posed in terms of identifying the transfer matrix of single-input multiple-output (SIMO) power system models using linear least-squares (LLS), where parallelism can be implemented through parallel execution of matrix multiplications using multiple processors or workers. Given the constraints of sequential communication and limited local memory, which may arise due to multiple applications running in the workers at the same time, a novel scheduling algorithm is proposed to enable flexible deadlines that meet these constraints. The scheduling algorithm minimizes the total time of execution under constraints, and can be solved via integer programming. The second problem is posed as a similar parallel algorithm for identifying a linearized state-variable (SV) model of a power system using both linear and nonlinear least-squares (NLS) in presence of scheduling. The performance of all the algorithms are studied via simulations of an IEEE 145-bus, 50-machine power system model, and compared with their centralized, non-parallel implementation.}, number={6}, journal={IEEE TRANSACTIONS ON POWER SYSTEMS}, author={Xue, Nan and Chakrabortty, Aranya}, year={2016}, month={Nov}, pages={4584–4594} }