Ping Ye

Ms. Ping Ye

orcid

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

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

Representative Publications

View All Publications
27 March 2024 in Remote Sensing
Verified
Journal Article
Open access
PDF file
PYPing Ye
Ping Ye
GXGang Xiao
Gang Xiao
JLJun Liu
Jun Liu
Vision–language tracking presents a crucial challenge in multimodal object tracking. Integrating language features and visual features can enhance target localization and improve the stability and accuracy of the tracking process. However, most existing fusion models in vision–language trackers simply concatenate visual and linguistic features without considering their semantic relationships. Such methods fail to distinguish the target’s appearance features from the background, particularly when the target changes dramatically. To address these limitations, we introduce an innovative technique known as multimodal features alignment (MFA) for vision–language tracking. In contrast to basic concatenation methods, our approach employs a factorized bilinear pooling method that conducts squeezing and expanding operations to create a unified feature representation from visual and linguistic features. Moreover, we integrate the co-attention mechanism twice to derive varied weights for the search region, ensuring that higher weights are placed on the aligned visual and linguistic features. Subsequently, the fused feature map with diverse distributed weights serves as the search region during the tracking phase, facilitating anchor-free grounding to predict the target’s location. Extensive experiments are conducted on multiple public datasets, and our proposed tracker obtains a success score of 0.654/0.553/0.447 and a precision score of 0.872/0.556/0.513 on OTB-LANG/LaSOT/TNL2K. These results are satisfying compared with those of recent state-of-the-art vision–language trackers.

5 August 2024 in IEEE Transactions on Instrumentation and Measurement
Verified
Journal Article
Closed access
PYPing Ye
Ping Ye
GXGang Xiao
Gang Xiao
JLJun Liu
Jun Liu
Multimodal visual object tracking (VOT) starts to replace a single modality by providing more complementary details of the target. In RGB plus depth (RGBD) tracking tasks, prevailing trackers extract RGB and depth features with separate feature extraction backbones in parallel structures and fuse the features for target prediction, where RGB is treated as the dominant modality, while the depth information are the assistant one. Their oversight lies in neglecting the distinct information-carrying capacities of each modality, and the fact that RGB and depth features contribute dynamically and variably to the tracking process. In response to these challenges, we have introduced a unified RGBD tracking network called AMATrack, which is designed to simultaneously extract features and investigate interactions between the individual RGB and depth images. Initially, the depth image is integrated with the RGB image to form the fused template and search region separately, ensuring maximal preservation of information from both modalities. Subsequently, we introduce an asymmetric mixed attention (AMA) module within the encoder layers. This module incorporates self-attention for maintaining intramodality features and cross-attention for minimizing discrepancies in intermodality features. Simultaneously, it removes correlations with nontarget patches to prevent adverse effects from background areas. To address the additional computational load during training and inference, we have customized a target token pruning (TTP) technique dubbed TTP. This strategy filters out patch tokens least relevant to the target before generating the final search region, thereby enhancing computational efficiency. In conclusion, our proposed AMATrack attains an F-score of 61.8% and 75.8% on the DepthTrack dataset and CDTB dataset, respectively, while operating at a speed of 73.01 frames/s (FPS). This performance surpasses that of the leading RGBD trackers in meeting real-time demands. These outcomes demonstrate the efficacy and superiority of our proposed network.