Search

Your search keyword '"Zhonglong Zheng"' showing total 264 results
Start Over Author "Zhonglong Zheng"

Refine your results

more »

more »

more »

more »
264 results on '"Zhonglong Zheng"'

85. SmokerViT: A Transformer-Based Method for Smoker Recognition.

Author

Khan, Ali, Khan, Somaiya, Hassan, Bilal, Khan, Rizwan, and Zhonglong Zheng

Subjects
TRANSFORMER models, DEEP learning, CONVOLUTIONAL neural networks, SMOKING bans, FALSE alarms, SMOKING cessation, POISONS, IMAGE representation
Abstract

Smoking has an economic and environmental impact on society due to the toxic substances it emits. Convolutional Neural Networks (CNNs) need help describing low-level features and can miss important information. Moreover, accurate smoker detection is vital with minimum false alarms. To answer the issue, the researchers of this paper have turned to a self-attention mechanism inspired by the ViT, which has displayed state-of-the-art performance in the classification task. To effectively enforce the smoking prohibition in non-smoking locations, this work presents a Vision Transformer-inspired model called SmokerViT for detecting smokers. Moreover, this research utilizes a locally curated dataset of 1120 images evenly distributed among the two classes (Smoking and NotSmoking). Further, this research performs augmentations on the smoker detection dataset to have many images with various representations to overcome the dataset size limitation. Unlike convolutional operations used in most existing works, the proposed SmokerViT model employs a self-attention mechanism in the Transformer block, making it suitable for the smoker classification problem. Besides, this work integrates the multi-layer perceptron head block in the SmokerViT model, which contains dense layers with rectified linear activation and linear kernel regularizer with L2 for the recognition task. This work presents an exhaustive analysis to prove the efficiency of the proposed SmokerViT model. The performance of the proposed SmokerViT performance is evaluated and compared with the existing methods, where it achieves an overall classification accuracy of 97.77%, with 98.21% recall and 97.35% precision, outperforming the state-of-the-art deep learning models, including convolutional neural networks (CNNs) and other vision transformer-based models. [ABSTRACT FROM AUTHOR]

Published
2023
Full Text
View/download PDF

93. Asymmetric Correlation Quantization Hashing for Cross-Modal Retrieval

Author

Jie Yang, Masoumeh Zareapoor, Zhonglong Zheng, and Lu Wang

Subjects
Pointwise, Optimization problem, Computer science, Hash function, Computer Science Applications, Data point, Semantic similarity, Signal Processing, Media Technology, Binary code, Pairwise comparison, Electrical and Electronic Engineering, Quantization (image processing), Algorithm
Abstract

In recent years, cross-modal hashing (CMH) has attracted considerable attention due to its ability to learn across different modalities and its high efficiency for similarity retrieval applications. This procedure is computationally inexpensive when dealing with large-scale multi-modalities datasets. However, they do not form the ideal representative model to fully exploit multi-modal data's underlying properties despite their successful performance. We identify that: (i) most CMH models in their current forms transform the real data points into discrete compact binary codes, which can limit their ability to prevent the loss of important information and thereby produce suboptimal results. (ii) the discrete-binary constraint model is hard to implement, and relaxing the binary constraints is a common property in most existing methods, which often leads to significant quantization errors. (iii) handling the CMH in a symmetry domain leads to a complex and inefficient optimization problem. This paper addresses the above challenges and proposes a novel Asymmetric Correlation Quantization Hashing (ACQH) method. ACQH learns a projection matrix for each heterogeneous modality to map the data point into a low-dimensional semantic space and constructs a compositional quantization to generate hash codes, using the pairwise semantic similarity preservation and the pointwise label regression. As a specific instantiation of our model, we use discrete iterative optimization to obtain the unified hash codes across different modalities. Extensive experiments show that ACQH outperforms state-of-the-art methods on several diverse datasets.

Published
2022

Catalog

Books, media, physical & digital resources
See catalog results