Zongyi Li

Zongyi Li

Auto-generated profile, based on publicly available information

Zongyi Li

Huazhong University of Science and Technology, China

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
2 February 2024 in IEEE Transactions on Biometrics, Behavior, and Identity Science
Verified
Journal Article
Closed access
HZHanqing Zheng
Hanqing Zheng
YSYuxuan Shi
Yuxuan Shi
HLHefei Ling
Hefei Ling
ZLZongyi Li
Zongyi Li
RWRunsheng Wang
Runsheng Wang
PLPing Li
Ping Li
Occluded person re-identification is a challenging task due to various noise introduced by occlusion. Previous methods utilize body detectors to exploit more clues which are overdependent on accuracy of detection results. In this paper, we propose a model named Cascade Transformer Reasoning Embedded by Uncertainty Network (CTU) which does not require external information. Self-attention of the transformer models long-range dependency to capture difference between pixels, which helps the model focus on discriminative information of human bodies. However, noise such as occlusion will bring a high level of uncertainty to feature learning and makes self-attention learn undesirable dependency. We invent a novel structure named Uncertainty Embedded Transformer (UT) Layer to involve uncertainty in computing attention weights of self-attention. Introducing uncertainty mechanism helps the network better evaluate the dependency between pixels and focus more on human bodies. Additionally, our proposed transformer layer generates an attention mask through Cascade Attention Module (CA) to guide the next layer to focus more on key areas of the feature map, decomposing feature learning into cascade stages. Extensive experiments over challenging datasets Occluded-DukeMTMC, P-DukeMTMC, etc., verify the effectiveness of our method.

17 July 2023 in IEEE Transactions on Computational Social Systems
Verified
Journal Article
Closed access
FZFengfan Zhou
Fengfan Zhou
HLHefei Ling
Hefei Ling
YSYuxuan Shi
Yuxuan Shi
JCJiazhong Chen
Jiazhong Chen
ZLZongyi Li
Zongyi Li
PLPing Li
Ping Li
Face recognition (FR) models can be easily fooled by adversarial examples, which are crafted by adding imperceptible perturbations on benign face images. The existence of adversarial face examples poses a great threat to the security of society. To build a more sustainable digital nation, in this article, we improve the transferability of adversarial face examples to expose more blind spots of the existing FR models. Though generating hard samples has shown its effectiveness in improving the generalization of models in training tasks, the effectiveness of using this idea to improve the transferability of adversarial face examples remains unexplored. To this end, based on the property of hard samples and the symmetry between training tasks and adversarial attack tasks, we propose the concept of hard models, which have similar effects as hard samples for adversarial attack tasks. Using the concept of hard models, we propose a novel attack method called beneficial perturbation feature augmentation attack (BPFA), which reduces the overfitting of adversarial examples to surrogate FR models by constantly generating new hard models to craft the adversarial examples. Specifically, in the backpropagation, BPFA records the gradients on preselected feature maps and uses the gradient on the input image to craft the adversarial example. In the next forward propagation, BPFA leverages the recorded gradients to add beneficial perturbations on their corresponding feature maps to increase the loss. Extensive experiments demonstrate that BPFA can significantly boost the transferability of adversarial attacks on FR.

25 October 2023 in IEEE Transactions on Multimedia
Verified
Journal Article
Closed access
RWRunsheng Wang
Runsheng Wang
YSYuxuan Shi
Yuxuan Shi
HLHefei Ling
Hefei Ling
ZLZongyi Li
Zongyi Li
CZChengxin Zhao
Chengxin Zhao
PLPing Li
Ping Li
Existing methods combining skeleton and silhouette representations demonstrate explicit effectiveness for gait recognition. However, current related methods simply combine the video-level representations of model-based skeleton data and gait silhouettes for retrieval. Therefore, diverse skeleton information is not fully exploited in existing related works: Firstly, the position and movement of bones are not clear from individual silhouettes. This indicates that the frame-level interaction between features of skeletons and silhouettes is critical, which is ignored by previous methods. Secondly, diverse part-level skeleton-guided gait features are not fully captured in existing related approaches. To solve the above issues, we present a novel framework with multi-level skeleton-guided refinement, including frame-level, part-level, and video-level skeleton-guided refinement, for comprehensive skeleton-aided gait representation learning. First, two modules are proposed for frame-level skeleton-guided refinement. Specifically, Visual Skeleton Enhanced Backbone (VSEB) is proposed to visually highlight the global and part-level skeleton regions for the feature of each silhouette frame. Moreover, Cross-Visual-Model Frame-level Interaction (CVMFI) is proposed to further transfer the model-based skeleton information to features of the visual modalities. Secondly, part-level visual and model-based skeleton features are utilized to refine the final gait representation. Concretely, in VSEB, Part Skeleton Enhance Network (PSEN) is proposed to visually enhance the position and movement of part-level skeletons. In addition, Semantic Part Pooling (SPP) is proposed for capturing the model-based skeleton features of different semantic parts. Finally, as the video-level skeleton-guided refinement, multimodal video-level features are combined to boost the final recognition performance. Extensive experimental results on prevailing datasets demonstrate that our approach outperforms most existing methods, including the skeleton-aided multi-modal methods. With the multi-level refinement guided by the skeleton modalities, the framework is expected to provide a deeper understanding of skeleton-aided gait recognition.

6 January 2025 in IEEE Transactions on Biometrics, Behavior, and Identity Science
Verified
Journal Article
Closed access
BLBoyuan Liu
Boyuan Liu
XZXin Zhang
Xin Zhang
HLHefei Ling
Hefei Ling
ZLZongyi Li
Zongyi Li
RWRunsheng Wang
Runsheng Wang
PLPing Li
Ping Li
Deep synthesis multimedia content, especially human face manipulation poses a risk of visual and auditory confusion, highlighting the call for generalized face forgery detection methods. In this paper, we propose a novel method for fake sample synthesis, along with a dual auto-encoder network for generalized deepfake detection. First, we delve into the texture discrepancy between tampered and unperturbed regions within forged images and impose models to learn such features by adopting Augmentation Inside Masks (AIM). It is capable of sabotaging the texture consistency within a single real image and generating textures that are commonly seen in fake images. It is realized by exhibiting forgery clues of discrepancy in noise patterns, colors, resolutions, and especially the existence of GAN (Generative Adversarial Network) features, including GAN textures, deconvolution traces, GAN distribution, etc. To the best of our knowledge, this work is the first to incorporate GAN features in fake sample synthesizing. The second is that we design a Bone-shaped dual auto-encoder with a powerful image texture filter bridged in between to aid forgery detection and localization in two streams. Reconstruction learning in the color stream avoids over-fitting in specific textures and imposes learning color-related features. Moreover, the GAN fingerprints harbored within the output image can be in furtherance of AIM and produce texture-discrepant samples for further training. The noise stream takes input processed by the proposed texture filter to focus on noise perspective and predict forgery region localization, subjecting to the constraint of mask label produced by AIM. We conduct extensive experiments on multiple benchmark datasets and the superior performance has proven the effectiveness of AIM-Bone and its advantage against current state-of-the-art methods. Our source code is available at https://github.com/heart74/AIM-Bone.git.

21 November 2025 in ACM Transactions on Multimedia Computing, Communications, and Applications
Verified
Journal Article
Open access
SHShijuan Huang
Shijuan Huang
ZLZongyi Li
Zongyi Li
HLHefei Ling
Hefei Ling
JLJianbo Li
Jianbo Li
Text-based person search aims to retrieve specific individuals from an extensive image gallery using textual queries. Recent approaches have delved into aligning global and part features in both text and image modalities, yielding substantial improvements. However, these methods often overlook intra-modality instance relations and uncertainties inherent in text-based person search. In response to these challenges, we propose the Cross-Modality Relation and Uncertainty Exploration (CRUE) method to model the relations and uncertainties in the matching procedure. To alleviate the strict alignment issues arising from hard labels in the original contrastive loss, an Intra-Modality Relation Exploration (IRE) module is introduced. This module is specifically designed to smooth hard-matching relations by modeling intra-modality similarity. Additionally, to address uncertain matching problems stemming from many-to-many relations, we propose a novel Uncertainty-Guided Modeling (UGM) module. This module is specifically designed to handle weak and noise-matched image–text pairs by modeling features as distributions, thereby alleviating instability and noise. Both the IRE and UGM modules effectively consider genuine intra-modality similarities and reduce the negative impact of uncertainties. Experimental results demonstrate significant improvements across three widely used person search datasets, thereby validating the efficacy of the CRUE method in enhancing text-based person search. Our code will be available on GitHub at https://github.com/ShijuanHuang/CRUE .

22 January 2024 in ACM Transactions on Multimedia Computing, Communications, and Applications
Verified
Journal Article
Open access
ZLZongyi Li
Zongyi Li
YSYuxuan Shi
Yuxuan Shi
HLHefei Ling
Hefei Ling
JCJiazhong Chen
Jiazhong Chen
BLBoyuan Liu
Boyuan Liu
CZChengxin Zhao
Chengxin Zhao
Unsupervised object Re-ID aims to learn discriminative identity features from a fully unlabeled dataset to solve the open-class re-identification problem. Satisfying results have been achieved in existing unsupervised Re-ID methods, primarily trained with pseudo-labels created by feature clustering. However, the viewpoint variation of objects is the key challenge, introducing noisy labels in the clustering process. To address this problem, a novel viewpoint disentangling and generation framework (VDG) is proposed to learn viewpoint-invariant ID features, including a disentangling and generation module, as well as a contrastive learning module. First, we design an ID encoder to map the viewpoint and identity features into the latent space. Second, a generator is used to disentangle view features and synthesize images with different orientations. Especially, the well-trained encoder serves as a pre-trained feature extractor in the contrastive learning module. Third, a viewpoint-aware loss and a class-level loss are integrated to facilitate contrastive learning between original and novel views. The generation of novel view images and the application of viewpoint-aware contrastive loss mutually assist model learning viewpoint-invariant ID features. Extensive experiments on Market-1501, DukeMTMC, MSMT17, and VeRi-776 demonstrate the effectiveness of the proposed VDG framework, as well as its superiority over the existing state-of-the-art approaches. The VDG model also demonstrates high quality in the image generation tasks.