作者:
Grina WiemDouik AliUniversity of Sousse
Networked Objects Control and Communication Systems Laboratory National School of Engineers of Sousse Sousse Technology Center Sahloul Belt Road 4054 University of Monastir National Engineering School of Monastir Rue Ibn Jazzar Monastir 5035 Tunisia University of Sousse
Networked Objects Control and Communication Systems Laboratory National School of Engineers of Sousse Sousse Technology Center Sahloul Belt Road 4054 Tunisia
The performance of recognition systems can be significantly affected by various factors, with facial expression poses and lighting changes being the main confounding factors. In order to minimize their impact, we prop...
详细信息
The performance of recognition systems can be significantly affected by various factors, with facial expression poses and lighting changes being the main confounding factors. In order to minimize their impact, we propose an intriguing method in this study that enables the generation of high-quality images specifically tailored to the target domain. Our objective is to utilize disentangled representation to effectively model the decomposition of data variations and generate neutral facial expression images with frontal posture and adaptive illumination. To achieve this, we incorporate 3D priors in the adversarial learning during the training process, simulating the generation of an analytical 3D face deformation as well as rendering operations. Additionally, we employ contrastive learning to control the disentanglement of the generated faces while preserving the essential properties of facial features. This technique enables us to learn an embedding space in which similar data samples are represented closely, while distinct samples are kept far apart from each other. Furthermore, we conduct an analysis of the learned latent space and introduce several other significant properties that enhance the reinforcement of factor disentanglement. These properties include an imitation learning algorithm, which facilitates the acquisition of meaningful patterns and characteristics.
This work proposes a novel generative model, FF-GAN (Frontal Face Generative Adversarial Network), for generating high-quality and diverse frontal faces. FF-GAN utilizes contrastive learning to effectively learn the u...
详细信息
This work proposes a novel generative model, FF-GAN (Frontal Face Generative Adversarial Network), for generating high-quality and diverse frontal faces. FF-GAN utilizes contrastive learning to effectively learn the underlying representation of frontal faces from unlabeled image datasets. This approach allows the model to capture the essential characteristics of frontal faces without the need for explicit pose annotations. We evaluate the performance of FF-GAN using established metrics like FID (Fréchet Inception Distance), IS (Inception Score), and SSIM (Structural Similarity Index measure). The results demonstrate that FF-GAN achieves superior performance compared to existing methods, generating highly realistic and visually appealing frontal faces with exceptional structural coherence. This research contributes to the field of facial image generation by introducing an effective unsupervised learning approach based on contrastive learning for generating high-quality frontal faces.
To address the problem of the unbalanced load and optimize the traffic distribution of large-scale low earth orbit (LEO) satellite networks, this paper proposes a load-balancing routing algorithm for LEO satellite net...
详细信息
The need for better search and rescue capabilities has led to the increased demand for collaborative aerial-ground multi-robot deployments. However, most existing solutions require high communication bandwidth or bulk...
详细信息
The need for better search and rescue capabilities has led to the increased demand for collaborative aerial-ground multi-robot deployments. However, most existing solutions require high communication bandwidth or bulky sensor equipment that is not suitable especially for aerial robots. To make the execution of these tasks more efficient, this paper proposes an angle-specified heterogeneous leader-follower formation framework consisting of unmanned aerial vehicles (UAVs) flying in a 3D space and ground robots moving in a 2D plane. The UAVs are controlled to maintain a desired angle-specified formation with the first three UAVs as the leaders forming a triangular shape and the remaining UAVs as the followers. The follower UAVs only need direction measurements to track the leader UAVs. For the ground robots, two leader robots track the UAV group and determine the orientation and scale of the ground formation, while the follower robots track the leader robots using only direction measurements. The proposed heterogeneous formation framework is energy-efficient as most robots only require low-cost and lightweight direction measurements. The stability of the proposed formation control algorithms is proved and validated through various physical application experiments. IEEE
Compared with traditional encryption technology, physical layer security technology uses channel noise to realize the key with the same length as plaintext. Even if the eavesdropper has strong cracking computing power...
详细信息
Artificial noise, as a physical layer security technique, differs from traditional communication security mechanisms based on cryptography as it cannot be easily cracked by rapidly developing computational power. The ...
详细信息
This paper proposes a random linear network coding method tailored for data transmission within Low Earth Orbit (LEO) constellation networks to address the challenges of long delay and high bit error rates (BER) in LE...
详细信息
ISBN:
(数字)9798350374513
ISBN:
(纸本)9798350374520
This paper proposes a random linear network coding method tailored for data transmission within Low Earth Orbit (LEO) constellation networks to address the challenges of long delay and high bit error rates (BER) in LEO satellite networks. The method employs a link quality-aware strategy, enabling dynamic adjustment of source-side redundancy based on the corresponding link quality to maximize network fault tolerance with minimal overhead. Validation using the OMNET++ discrete event simulation platform indicates that the proposed method performs well in LEO satellite networks of various scales and configurations, exhibiting superior performance in more extensive networks. In particular, in the 300-LEO and 200-LEO scenarios, the proposed method demonstrates a successful delivery rate improvement of 19.6% and 18.73%, respectively, compared to traditional single-path approaches. This method is deemed more suitable for application in dynamic satellite networks with large hop counts.
The timbre attributes are affected by multiple factors such as the time domain, frequency domain, time-frequency domain, and harmonic structure of the sound signal, and there is a certain degree of correlation and red...
详细信息
With the increasing shortage of wireless spectrum resources, it is urgent to monitor and utilize the limited wireless spectrum resources rationally. In this paper, a scattered wireless spectrum sensor network system i...
详细信息
To tackle challenges such as interference and poor accuracy of indoor positioning systems,a novel scheme based on ultra-wide bandwidth(UWB)technology is ***,we illustrate a distance measuring method between two UWB **...
详细信息
To tackle challenges such as interference and poor accuracy of indoor positioning systems,a novel scheme based on ultra-wide bandwidth(UWB)technology is ***,we illustrate a distance measuring method between two UWB ***,a Taylor series expansion algorithm is developed to detect coordinates of the mobile node using the location of anchor nodes and the distance between *** results show that the observation error under our strategy is within 15 cm,which is superior to existing *** final experimental data in the hardware system mainly composed of STM32 and DW1000 also confirms the performance of the proposed scheme.
暂无评论