YDYajun Ding

Yajun Ding

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Yajun Ding has a total of 34 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Dual-space intervention for mitigating bias in robust visual question answering

    by Runmin Wang, Xingdong Song, Zukun Wan, Han Xu, Congzhen Yu, Tianming Ma, Yajun Ding, Shengyou Qian


  • Adaptive bias learning via gradient-based reweighting and constrained pruning for robust Visual Question Answering

    by Zukun Wan, Runmin Wang, Xingdong Song, Juan Xu, Xiaofei Cao, Jielei Hei, Shengrong Yuan, Yajun Ding, Changxin Gao

  • LTDNet: A Lightweight Text Detector for Real-Time Arbitrary-Shape Traffic Text Detection

    by Runmin Wang, Yanbin Zhu, Ziyu Zhu, Lingxin Cui, Zukun Wan, Anna Zhu, Yajun Ding, Shengyou Qian, Changxin Gao, Nong Sang

    Traffic text detection plays a vital role in understanding traffic scenes. Traffic text, a distinct subset of natural scene text, faces specific challenges not found in natural scene text detection, including false alarms from non-traffic text sources, such as roadside advertisements and building signs. Existing state-of-the-art methods employ increasingly complex detection frameworks to pursue higher accuracy, leading to challenges with real-time performance. In response to this issue, we propose a real-time and efficient traffic text detector named LTDNet, which strikes a balance between accuracy and real-time capabilities. LTDNet integrates three essential techniques to address these challenges effectively. First, a cascaded multilevel feature fusion network is employed to mitigate the limitations of lightweight backbone networks, thereby enhancing detection accuracy. Second, a lightweight feature attention module is introduced to enhance inference speed without compromising accuracy. Finally, a novel point-to-edge distance vector loss function is proposed to precisely localize text instance boundaries within traffic contexts. The superiority of our method is validated through extensive experiments on five publicly available datasets, demonstrating its state-of-the-art performance. The code will be released at https://github.com/runminwang/LTDNet.


  • FSANet: Feature shuffle and adaptive channel attention network for arbitrary shape scene text detection

    by Juan Xu, Runmin Wang, Jielei Hei, Xiaofei Cao, Zukun Wan, Congzhen Yu, Yajun Ding, Changxin Gao, Shengyou Qian


  • TTDNet: An End-to-End Traffic Text Detection Framework for Open Driving Environments

    by Runmin Wang, Yanbin Zhu, Hua Chen, Zhenlin Zhu, Xiangyu Zhang, Yajun Ding, Shengyou Qian, Changxin Gao, Li Liu, Nong Sang

    Traffic text detection is crucial for traffic scene understanding in intelligent transportation systems (ITS). Although natural scene text detection has been extensively studied, yielding noteworthy results, little research has focused on traffic text detection. Traffic text, as a special type of natural scene text, faces not only the general challenges of natural scene text detection but also the significant impact of false alarms from non-traffic texts on system performance. In light of these challenges, we propose an end-to-end traffic text detection framework that can effectively detect traffic text captured by in-vehicle cameras in various driving scenarios. The key contributions of our proposed approach are: (1) an Image Enhancement Module designed to remove fog and enhance low-quality images; (2) a plug-and-play Text Feature Enhancement Module; (3) a Joint Loss Function; and (4) the creation of an open driving environment traffic text dataset (named ODETT-3000) containing various traffic environments. Comprehensive experimental studies have verified that our method achieves state-of-the-art performance on traffic text datasets such as CTST-1600, TPD, and ODETT-3000, as well as promising results on public natural scene text datasets such as MSRA-TD500, ICDAR 2015, and CTW 1500, thereby showcasing the superior performance and adaptability of our method. The code and our dataset (ODETT-3000) will be available at https://github.com/yanbin-zhu/TTDNet.


  • CR2-Net: Component Relationship Reasoning Network for Traffic Text Detection

    by Runmin Wang, Jielei Hei, Minghao Liu, Zukun Wan, Juan Xu, Xiaofei Cao, Xuan He, Yajun Ding, Changxin Gao, Nong Sang

    Traffic text, as a special type of natural scene text, plays an important role in intelligent transportation. Apart from the challenges faced by general natural scene text detection, the presence of a large amount of non-traffic text which leads to false alarms further complicates the problem of traffic text detection, e.g., billboards, building signs, etc. In this work, we infer text component relationships via a component relationship reasoning network (CR2-Net) for traffic text detection. The method contains main contributions as follows: (1) explored a GCN-based component-relational reasoning traffic text detection network by leveraging global contextual information and reasoning about intrinsic connections between text components to improve traffic text detection performance. (2) designed a Group Deformable Convolution Network (GDCN) and adaptively adjusting the convolutional manner to enhance the feature extraction capability of the network for text regions. (3) proposed a Fusion Attention Module (FAM) to enhance the network adaptive ability by assigning feature map weights to different channels. To demonstrate the effectiveness, we evaluate CR2-Net on four standard benchmarks. Experimental results demonstrate that our method achieves remarkable performance on two traffic text datasets, i.e., CTST-1600, and TPD. Especially, our results on TPD dataset surpass all previously reported results to the best of our knowledge. Additionally, our approach exhibits competitive results on general natural scene text datasets, such as ICDAR 2015, and SCUT-CTW1500, thereby fully demonstrating the superiority and adaptability of our method. The code will be publicly available at https://github.com/runminwang/CR2-Net .


  • DIMGNet: A Transformer-Based Network for Pedestrian Reidentification With Multi-Granularity Information Mutual Gain

    by Runmin Wang, Zhenlin Zhu, Yanbin Zhu, Hua Chen, Yongzhong Liao, Ziyu Zhu, Yajun Ding, Changxin Gao, Nong Sang

    Pedestrian reidentification (ReID) is a challenging task that involves identifying and retrieving specific pedestrians across different cameras and scenes. This problem has significant implications for security surveillance, and has thus received substantial attention in recent years. However, traditional convolutional neural networks (CNNs) have limited receptive fields and cannot capture global information. Moreover, transformer networks, which excel in long-range feature capture, are prone to accuracy degradation due to loss of details. To address these limitations, we propose a transformer-based pedestrian ReID network with double-branch information mutual gain (DIMGNet), which leverages hierarchical parallel levels to support multi-granularity feature information mutual gain. Our model also incorporates an auxiliary camera information (ACI) module to improve feature representation ability. We further embed a cross-attention mechanism into the architecture to enhance mutual gain between multi-granularity features and improve feature discrimination. Finally, we introduce a shuffling technique to increase the robustness of the extracted features. We evaluate the proposed method on several benchmark datasets, including Market-1501[1], MSMT17[2], DukeMTMC-reID [3], and Occluded-Duke [4], achieving $mAP$ values of 90.7%, 68.4%, 83.7%, and 60.6%, respectively. Our method outperforms most state-of-the-art methods, demonstrating the effectiveness of our method. The code will be publicly released at https://github.com/ZhenlinZhu/DIMGNet .


  • Open Access

    FENet: Feature Enhancement Network for Arbitrary Direction Text Detection

    by Runmin Wang, Yingying Liu, Hua Chen, Yanbin Zhu, Guilin Zhu, Minghao Liu, Chang Han, Yajun Ding

    With the advent of deep learning, scene text detection has gained significant prominence, finding applications in diverse fields such as autonomous driving, and scene text translation. Existing frameworks often encode text candidates independently, employing sequential deep convolutional methods (e.g., ResNet, VGGNet). These methods grapple with several challenges, including the variability of text shapes, the presence of intricate backgrounds, and limitations in retaining relatively long contextual information. In contrast, our proposed network, denominated as the Feature Enhancement Network (FENet), adeptly addresses arbitrary direction text detection through the integration of attention mechanisms and Bi-directional Long Short-Term Memory (BLSTM). Our method boasts three pivotal attributes: (i) Recurrent strengthening of feature maps via the amalgamation of an attention mechanism-based Convolutional Block Attention Module (CBAM) and a BLSTM-based sequence feature extraction module. (ii) Modification of the constituent elements of the bottleneck layer within the ResNet architecture to facilitate micro-batch training. (iii) Augmentation of network learnability through the incorporation of residual network structures within the BLSTM-based sequence feature extraction module. We conducted extensive experiments on five publicly available datasets, namely ICDAR 2013, ICDAR 2015, MSRA-TD500, SCUT-CTW1500, and ICDAR2017-MLT. The experimental results demonstrate the efficacy of our proposed method. We will release code to facilitate community research at the website: https://github.com/lyy0117.

  • Open Access

    FTN–VQA: MULTIMODAL REASONING BY LEVERAGING A FULLY TRANSFORMER-BASED NETWORK FOR VISUAL QUESTION ANSWERING

    by Runmin Wang, Weixiang Xu, Yanbin Zhu, Zhenlin Zhu, Hua Chen, Yajun Ding, Jinping Liu, Changxin Gao, Nong Sang

    Visual Question Answering (VQA) is a multimodal task, which requires understanding the information in the natural language questions and paying attention to the useful information in the images. So far, the solution of VQA tasks can be divided into grid-based methods and bottom-up-based methods. The grid-based method directly extracts the semantic features of the image by leveraging the convolution neural network (CNN), so it has a praiseworthy computational efficiency, but the global convolution feature ignores the information of the key area and causes the performance bottleneck. The bottom-up-based method needs to detect potentially problem-related objects by using some object detection frameworks, e.g. Faster RCNN, so it has better performance, but the computational efficiency is reduced due to the computation of Region Proposal Network (RPN) and Non-Maximum Suppression (NMS). Based on the aforementioned reasons, we propose a fully transformer-based network (FTN) that can maintain a balance between computational efficiency and accuracy, which can be trained end-to-end and consists of three components: question module, image module, and fusion module. Meanwhile, the image module and the question module are visualized to explore the operating rules of the transformer. The experiment results demonstrate that the FTN can focus on key information and objects in the question module and the image module, and our single model has reached 69.01% accuracy on the VQA2.0 dataset, which is superior to the grid-based methods. Although FTN fails to surpass a few state-of-the-art bottom-up-based methods, the FTN has obvious advantages in computational efficiency. The code will be released at https://github.com/weixiang-xu/FTN-VQA.git.