ZSZiwen Sun

Ziwen Sun

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Ziwen Sun has a total of 9 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    A Progressive-Assisted Object Detection Method Based on Instance Attention

    by Ziwen Sun, Zhizhong Xi, Hao Li, Chong Ling, Dong Chen, Xiaoyan Qin

    Overcoming the high cost of self-attention operation in Transformer-based object detection methods and improving the detection accuracy of small objects is one of the difficulties in the field of object detection research. This paper designs a progressive assisted object detection method PaoDet based on Transformer, which uses common feature extraction backbone such as Resnet and ViT to extract multi-scale features of the input image, and uses RPN(Region Proposal Network) to extract proposals of different scales; Subsequently, a progressive modeling approach was adopted to perform self-attention and cross-attention operations on proposals of different scales from large to small, achieving feature interaction between instances, ensuring high detection efficiency and low computational complexity. During the training process, each layer of the network has certain generalization ability for detecting adjacent scale objects under the supervision of a dynamic scale division method. Compared with state-of-the-art object detection methods on COCO and UAVDT datasets, the effectiveness and superiority of the proposed method were demonstrated.

  • Open Access

    Siamese Contour Segmentation Network for Multi-State Object Tracking

    by Hao Li, Ziwen Sun, Chong Ling, Chao Xu

    Tracking multiple states of the target simultaneously is currently a research hotspot in object tracking. Existing methods obtain the initial bounding box with multi-scale search, anchor-based regression, or anchor-free regression. Then, each pixel within bounding box is classified and the mask of target is used to fit a fine bounding box for accurate tracking. However, their computation complexity is high and the accuracy of mask and fine bounding box is limited by initial bounding box. Based on SiamMask, the Siamese contour segmentation network (SiamCS) is proposed for multi-state object tracking to address these issues. This end-to-end network eliminates need to pre-define anchor based on prior knowledge, and reduces hyperparameters. With SiamCS, multi-state object tracking method formulates object tracking as region classification and contour regression to obtain bounding box, contour, and mask of target at the same time. Moreover, according to geometric meaning of definite integral and difficulty of sample, difficulty sensitive contour-intersection over union loss function is proposed to solve the problem of independent regression of contour parameters. Extensive experiments on OTB100 (92.5%Precision), UAV123 (83.5%Precision), LaSOT (64.8%AUC), TrackingNet (81.5%AUC), GOT-10K (66.3%AO), and VOT2020 (53.5%EAO) show that SiamCS outperforms many state-of-the-art trackers and achieves leading performance.


  • Open Access

    A Lightweight Object Tracking Method Based on Transformer

    by Ziwen Sun, Chuandong Yang, Chong Ling

    Transformer is widely used in the field of object tracking by virtue of its powerful global modeling capability, but the existing Transformer-based object tracking method is slow due to the large number of parameters, which limits the use of the algorithm on mobile smart devices, in order to reduce the number of parameters to further improve the tracking speed, a lightweight object tracking based on Transformer is proposed. Transformer-based lightweight object tracking method is proposed. First, in the feature extraction part, the first three layers of Swin-Transformer network are used to extract deep features from the input initial template, dynamic template, and search area, respectively. Second, in order to make full use of the initial template information, the extracted features are feature-enhanced with the help of the cross-attention module, and the features after the enhancement of the initial template and the dynamic template are sparsified by the sparsification module. Then, the fused features are fed into the simplified encoder and decoder modules for feature fusion. Finally, the output features go through a bounding box prediction header to predict the tracked object and a quality scoring header to determine whether to update the dynamic template. The algorithm is experimented on the LaSOT dataset, and the tracking success rate reaches 70.8% and the tracking speed reaches 78.7 FPS. The tracking speed is improved by significantly reducing the number of parameters while guaranteeing the tracking accuracy, which proves the effectiveness of the lightweight tracking method.

  • Adaptive Local Reinforce Attention Network for Few-shot Learning

    by Minghao Yan, Qinxin Lu, Ziwen Sun

    In many real-world application scenarios, limited to the economic, time and labor factors, it is not feasible to collect a large-scale labeled dataset. Therefore, researchers proposed few-shot learning methods that do not rely on large-scale data training. However, existing few-shot learning methods only use the global feature map of an image to predict its class, ignoring the role of information of local details with category characteristics in the image for classification. In this paper, we propose an adaptive local reinforce attention network (ALRANet) that can capture and strengthen information of local details. By dividing the learned knowledge, different kinds of learned knowledge are used to guide the capture and enhancement of local details in the new sample, and the content of learned knowledge is dynamically updated according to the learning performance. Extensive experiments are conducted on two datasets, and the proposed ALRA model outperforms those existing model, demonstrating the effectiveness of our model.