Bi‐level deep mutual learning assisted multi‐task network for occluded person re‐identification

Wang, Yi; Wang, Liangbo; Zhou, Yu

doi:10.1049/ipr2.12688

Cited by 5 publications

(1 citation statement)

References 57 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In recent years, deep learning and intelligent algorithms has developed rapidly [50][51][52][60][61][62], most current approaches utilize deep neural networks to extract both global and local representation from individuals [14][15][16]. Zheng et al [15] treat each person as a separate class and designed a multi-class loss function to allow the network to extract discriminative global features.…”

Section: Related Work 21 General Person Re-idmentioning

confidence: 99%

Attention-Enhanced Controllable Disentanglement for Cloth-Changing Person Re-identification

Ge,

Yu,

Chen

et al. 2024

Preprint

View full text Add to dashboard Cite

Cloth-changing person re-identification (Re-ID) is an emerging research theme that aims at identifying individuals after clothing change. Many contemporary approaches focus on disentangling clothing features and solely employ clothing-unrelated parts for identification. However, the absence of ground truth poses a significant challenge to the disentanglement process, these methods may introduce unintended noise and degrade the overall performance. To mitigate this issue, we propose a novel framework, termed Attention-based Controllable Disentanglement Network (ACD-Net). In ACD-Net, we design an Attention-enhanced Disentanglement Branch (ADB) where human parsing masks are introduced to guide the separation of clothing features and clothing-unrelated features. Here, clothing-unrelated features are subdivided into unclothed body features and contour features, and we propose two novel attention mechanisms: Dynamic Interaction-Remote Aggregation Attention (DI-RAA) and Dynamic Interaction-Positional Relevance Attention (DI-PRA) to enhance the representations of these two features, respectively. Experimental results on PRCC, LTCC, DeepChange, and CCVID datasets demonstrate the superiority of our approach over the state-of-the-art methods. For the cloth-changing setting, the mAP of our network on PRCC, LTCC, and DeepChangedatasets are 59.5%, 22.6%, and 20.6%, and the Rank-1 are 60.6%, 45.5%, and 56.8%, respectively. In addition, our model also obtains 81.5% of mAP and 83.4% of Rank-1 on the video dataset CCVID. The code is available at: https://github.com/jk-love-ge/ACDNet.

show abstract

Section: Related Work 21 General Person Re-idmentioning

confidence: 99%

Attention-Enhanced Controllable Disentanglement for Cloth-Changing Person Re-identification

Ge,

Yu,

Chen

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

Harmonious Mutual Learning for Facial Emotion Recognition

Gan,

Xu,

Xia

et al. 2024

Neural Process Lett

View full text Add to dashboard Cite

Facial emotion recognition in the wild is an important task in computer vision, but it still remains challenging since the influence of backgrounds, occlusions and illumination variations in facial images, as well as the ambiguity of expressions. This paper proposes a harmonious mutual learning framework for emotion recognition, mainly through utilizing attention mechanisms and probability distributions without utilizing additional information. Specifically, this paper builds an architecture with two emotion recognition networks and makes progressive cooperation and interaction between them. We first integrate self-mutual attention module into the backbone to learn discriminative features against the influence from emotion-irrelevant facial information. In this process, we deploy spatial attention module and convolutional block attention module for the two networks respectively, guiding to enhanced and supplementary learning of attention. Further, in the classification head, we propose to learn the latent ground-truth emotion probability distributions using softmax function with temperature to characterize the expression ambiguity. On this basis, a probability distribution distillation learning module is constructed to perform class semantic interaction using bi-directional KL loss, allowing mutual calibration for the two networks. Experimental results on three public datasets show the superiority of the proposed method compared to state-of-the-art ones.

show abstract