ZLZiyang Li

Ziyang Li

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Ziyang Li has a total of 8 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Small Object Detection Based on Lightweight Feature Pyramid

    by Ziyang Li, Chenwei Guo, Guang Han

    In today’s information age, consumer drones have received more and more attention in the field of consumer electronics. Consumer drones are widely used by consumers in aerial photography, follow-up photography and other scenes. There are many researches that can be carried out in the perspective of UAV, and object detection in the perspective of UAV belongs to one of them. Driven by object detection technology, drones with intelligent perception capabilities can achieve efficient and flexible data collection capabilities. However, due to the large number, large scale variation and uneven distribution of small and medium-sized targets captured by UAV, the existing algorithms have high miss detection rate and error rate for small and medium-sized targets captured by UAV. In order to be better applied to the object detection task in the view of UAV, three modules are designed in this paper. The small object enhancement module, which consists of a feature downsampling layer and a convolution-free step layer, to better cope with the detection and classification tasks of small objects and low-resolution images. The lightweight upsampling feature pyramid structure proposed at the same time has a large receptive field, through the convolution kernel forecast block and the characteristic integration block, it makes the semantics more balanced and enhances the ability of the model to locate small objects. Finally, FocalMAE Loss which reduces the negative impact of low quality samples on the gradient and helps the model converge quickly. The proposed method achieves 8.1% improvement over the conventional method. By reducing the number of parameters and maintaining lightweight functions, it exhibits improved performance in detecting small objects in UAV images.


  • Small Object Detection Based on Microscale Perception and Enhancement-Location Feature Pyramid

    by Guang Han, Chenwei Guo, Ziyang Li, Haitao Zhao

    Due to the large number of small objects, significant scale variation, and uneven distribution in images captured by unmanned aerial vehicles (UAVs), existing algorithms have high rates of missing and false detections of small objects in drone images. A new object detection algorithm based on microscale perception and enhancement-location feature pyramid is proposed in this paper. The microscale perception module alternatives the original convolution module in backbone, changing the receptive field through two dilation branches with various dilation rates and an adjustment switch branch. To better match the size and shape of sampled targets, the weighted deformable convolution is employed. The enhancement-location feature pyramid module aggregates the features from each layer to obtain balanced semantic information and refines aggregated features to enhance their ability to represent features. Moreover, a bottom-up branch structure is added to utilize the property of lower-layer features being beneficial to locating small objects to enhance the localization ability for small objects. Additionally, by using specific image cropping and combining techniques, the target distribution of the training data is altered to make the model more sensitive to small objects and improving its robustness. Finally, a sample balance strategy is used in combination with Focal loss and a sample extraction control method to balance simple hard sample imbalance and the long-tail distribution of inter-class sample imbalance during training. Experimental results show that the proposed algorithm achieves a mean average precision of 35.9% on the VisDrone2019 dataset, which is a 14.2% improvement over the baseline Cascade RCNN and demonstrates better performance in detecting small objects in drone images. Compared with advanced algorithms in recent years, it also achieves state-of-the-art detection accuracy.


  • A Two-Stage Foveal Vision Tracker Based on Transformer Model

    by Guang Han, Jianshu Ma, Ziyang Li, Haitao Zhao

    With the development of Transformer visual models, attention-based trackers have shown highly competitive performance in the field of object tracking. However, in some tracking scenarios, especially those with multiple similar objects, the performance of existing trackers is often not satisfactory. In order to improve the performance of trackers in such scenarios, inspired by the fovea vision structure and its visual characteristics, this paper proposes a novel Foveal Vision Tracker (FVT). FVT combines the process of human eye fixation and object tracking, pruning based on the distance to the object rather than attention scores. This pruning method allows the receptive field of the feature extraction network to focus on the object, excluding background interference. FVT divides the feature extraction network into two stages: local and global, and introduces the Local Recursive Module (LRM) and the View Elimination Module (VEM). LRM is used to enhance foreground features in the local stage, while VEM generates circular fovea-like visual field masks in the global stage and prunes tokens outside the mask, guiding the model to focus attention on high-information regions of the object. Experimental results on multiple object tracking datasets demonstrate that the proposed FVT achieves stronger object discrimination capability in the feature extraction stage, improves tracking accuracy and robustness in complex scenes, and achieves a significant accuracy improvement with an AO of 72.6% on the GOT-10k dataset.


  • Text-to-Image Person Re-Identification Based on Multimodal Graph Convolutional Network

    by Guang Han, Min Lin, Ziyang Li, Haitao Zhao, Sam Kwong

    Text-to-image person re-identification (ReID) is a common subproblem in the field of person re-identification and image-text retrieval. Recent approaches generally follow the structure of a dual-stream network, extracting image and text features. There is no deep interaction between images and text in this approach, making it difficult for the network to learn a highly semantic feature representation. In addition, for both image data and text data, the feature extraction process is modeled in a regular way, such as using Transformer to extract sequence embeddings. However, this type of modeling disregards the inherent relationships among multimodal input embeddings. A more flexible approach to mining multimodal data, which uniformly treats the data as graphs, is proposed. In this way, the extraction and interaction of multimodal information are accomplished by means of messages passing between graph nodes. First, a unified multimodal feature extraction and fusion network is proposed based on the graph convolutional network, which enables the progression of multimodal information from ‘local’ to ‘global’. Second, an asymmetric multilevel alignment module, which focuses on more accurate ‘local’ information from a ‘global’ perspective, is proposed to progressively divide the multimodal information at each level. Last, a cross-modal representation matching strategy based on similarity distribution and mutual information is proposed to achieve cross-modal alignment. The proposed algorithm in this paper is simple and efficient, and the testing results on three public datasets (CUHK-PEDES, ICFG-PEDES and RSTPReID) show that it can achieve SOTA-level performance.