Yongjian Deng

Yongjian Deng

Auto-generated profile, based on publicly available information

Yongjian Deng

College of Computer Science, Beijing University of Technology, Beijing, China

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

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

Representative Publications

View All Publications
20 January 2023 in IEEE Transactions on Multimedia
Verified
Journal Article
Closed access
HLHai Liu
Hai Liu
CZCheng Zhang
Cheng Zhang
YDYongjian Deng
Yongjian Deng
BXBochen Xie
Bochen Xie
Tingting LiuTingting Liu
Tingting Liu
YLYou-Fu Li
You-Fu Li
Fine-grained bird image classification (FBIC) is not only meaningful for endangered bird observation and protection but also a prevalent task for image classification in multimedia processing and computer vision. However, FBIC suffers from several challenges, such as bird molting, complex background, and arbitrary bird posture. To effectively tackle these challenges, we present a novel invariant cues-aware feature concentration Transformer (TransIFC), which learns invariant and core information in bird images. To this end, two novel modules are proposed to leverage the characteristics of bird images, namely, the hierarchy stage feature aggregation (HSFA) module and the feature in feature abstraction (FFA) module. The HSFA module aggregates the multiscale information of bird images by concatenating multilayer features. The FFA module extracts the invariant cues of birds through feature selection based on discrimination scores. Transformer is employed as the backbone to reveal the long-dependent semantic relationships in bird images. Moreover, abundant visualizations are provided to prove the interpretability of the HSFA and FFA modules in TransIFC. Comprehensive experiments demonstrate that TransIFC can achieve state-of-the-art performance on the CUB-200-2011 dataset (91.0%) and the NABirds dataset (90.9%). Finally, extended experiments have been conducted on the Stanford Cars dataset to suggest the potential of generalizing our method on other fine-grained visual classification tasks.

14 November 2023 in IEEE Transactions on Image Processing
Verified
Journal Article
Open access
HLHai Liu
Hai Liu
CZCheng Zhang
Cheng Zhang
YDYongjian Deng
Yongjian Deng
Tingting LiuTingting Liu
Tingting Liu
Zhaoli ZhangZhaoli Zhang
Zhaoli Zhang
YLYou-Fu Li
You-Fu Li
Head pose estimation (HPE) is an indispensable upstream task in the fields of human-machine interaction, self-driving, and attention detection. However, practical head pose applications suffer from several challenges, such as severe occlusion, low illumination, and extreme orientations. To address these challenges, we identify three cues from head images, namely, critical minority relationships, neighborhood orientation relationships, and significant facial changes. On the basis of the three cues, two key insights on head poses are revealed: 1) intra-orientation relationship and 2) cross-orientation relationship. To leverage two key insights above, a novel relationship-driven method is proposed based on the Transformer architecture, in which facial and orientation relationships can be learned. Specifically, we design several orientation tokens to explicitly encode basic orientation regions. Besides, a novel token guide multi-loss function is accordingly designed to guide the orientation tokens as they learn the desired regional similarities and relationships. Experimental results on three challenging benchmark HPE datasets show that our proposed TokenHPE achieves state-of-the-art performance. Moreover, qualitative visualizations are provided to verify the effectiveness of the token-learning methodology.

12 August 2020 in IEEE Transactions on Image Processing
Verified
Journal Article
Closed access
HCHao Chen
Hao Chen
YDYongjian Deng
Yongjian Deng
YLYoufu Li
Youfu Li
THTzu-Yi Hung
Tzu-Yi Hung
GLGuosheng Lin
Guosheng Lin
Depth is beneficial for salient object detection (SOD) for its additional saliency cues. Existing RGBD SOD methods focus on tailoring complicated cross-modal fusion topologies, which although achieve encouraging performance, are with a high risk of over-fitting and ambiguous in studying cross-modal complementarity. Different from these conventional approaches combining cross-modal features entirely without differentiating, we concentrate our attention on decoupling the diverse cross-modal complements to simplify the fusion process and enhance the fusion sufficiency. We argue that if cross-modal heterogeneous representations can be disentangled explicitly, the cross-modal fusion process can hold less uncertainty, while enjoying better adaptability. To this end, we design a disentangled cross-modal fusion network to expose structural and content representations from both modalities by cross-modal reconstruction. For different scenes, the disentangled representations allow the fusion module to easily identify, and incorporate desired complements for informative multi-modal fusion. Extensive experiments show the effectiveness of our designs and a large outperformance over state-of-the-art methods.

6 January 2022 in IEEE Robotics and Automation Letters
Verified
Journal Article
Closed access
BXBochen Xie
Bochen Xie
YDYongjian Deng
Yongjian Deng
ZSZhanpeng Shao
Zhanpeng Shao
HLHai Liu
Hai Liu
YLYoufu Li
Youfu Li
Event cameras can perceive pixel-level brightness changes to output asynchronous event streams, and have notable advantages in high temporal resolution, high dynamic range and low power consumption for challenging vision tasks. To apply existing learning models on event data, many researchers integrate sparse events into dense frame-based representations which can work with convolutional neural networks directly. Although these works achieve high performance on event-based classification, their models need lots of parameters to process dense event frames which do not fit with the sparsity of event data. To utilize the sparse nature of events, we propose a voxel-wise graph learning model ( VMV-GCN ) for spatio-temporal feature learning on event streams. Specifically, we design the volumetric multi-view fusion module ( VMVF ) to extract spatial and temporal information from views of voxelized event data. Then we take representative event voxels as vertices and use a novel dual-graph construction strategy to connect them. By aggregating neighborhood information based on relationships of vertices, the proposed dynamic neighborhood feature learning module ( DNFL ) can capture discriminative spatio-temporal features on dynamically updated graphs. Experiments show that our method achieves state-of-the-art performance with low model complexity on event-based classification tasks, such as object classification and action recognition.

16 April 2021 in IEEE Transactions on Circuits and Systems for Video Technology
Verified
Journal Article
Closed access
YDYongjian Deng
Yongjian Deng
HCHao Chen
Hao Chen
YLYoufu Li
Youfu Li
Event-based object recognition has drawn increasing attention for event cameras’ distinguished advantages of low power consumption and high dynamic range. For this new modality, previous works based on customizing low-level descriptors are vulnerable to noise and with limited generalizability. Although recent works turn to design various deep neural networks to extract event features, they either suffer from data insufficiency to fully train the event-based model or fail to encode spatial and temporal cues simultaneously with their single view network. In this work, we address these limitations by proposing a multi-view attention-aware network, in which an event stream is projected to multi-view 2D maps to utilize well-trained 2D models and explore spatio-temporal complements. Besides, the attention mechanism is used to boost the complements in different streams for better joint inference. Comprehensive experiments show the large superiority of our model over state-of-the-art methods as well as the efficacy of our multi-view fusion framework for event data.

15 June 2020 in IEEE Robotics and Automation Letters
Verified
Journal Article
Closed access
YDYongjian Deng
Yongjian Deng
YLYoufu Li
Youfu Li
HCHao Chen
Hao Chen
Event cameras, with low power consumption, high temporal resolution, and high dynamic range, have been used increasingly in computer vision. These superior characteristics enable event cameras to perform low-energy and high-response object classification tasks in challenging scenarios. Nevertheless, specific encoding methods for event-based classification are required owing to the unconventional output of event cameras. Existing event-based encoding methods have focused on extracting semantic and motion information in event signals. However, two main problems exist in these methods: $(i)$ the motion information of event signals leads to mispredictions by the classifiers. $(ii)$ effective evaluation methods to validate the motion robustness of event-based classification models have yet to be proposed. In this work, we introduce an adaptive motion-agnostic encoder for event streams to address the first problem. The proposed encoder would allow us to extract indistinguishable semantic information from an object in different motion circumstances. In addition, we propose a novel motion inconsistency evaluation method to assess the motion robustness of the classification models. We apply our method to several benchmark datasets and evaluate it using motion consistency and inconsistency testing methods. Classification performance shows that our proposed encoder outperforms state-of-the-art methods by a large margin.