Joint face super-resolution and frontalization of non-frontal low-resolution faces is of significant importance for many face analysis applications. However, at the same time, it is a challenging task. In this paper, ...
详细信息
ISBN:
(数字)9798331510138
ISBN:
(纸本)9798331510145
Joint face super-resolution and frontalization of non-frontal low-resolution faces is of significant importance for many face analysis applications. However, at the same time, it is a challenging task. In this paper, a novel Guided Transformative Generative Adversarial Network (GT-GAN) is proposed to address the challenge of generating frontal high-resolution face images from non-frontal low-resolution face images. Existing methods usually treat super-resolution and frontalization as two separate tasks, leading to unsatisfactory processing results. GTG AN innovatively achieves both goals within a unified frame-work. Specifically, we propose Profile Attention Guidance (PAG) to generate landmark heatmaps that provide a priori pose information to the generative network and direct its attention to critical facial regions. In addition, an Attention Fusion Transformation Module (AFTM) is proposed to significantly improve the hallucination performance of the network by utilizing the attention mechanism and STNs, combined with sequential feature enhancement strategies. The experimental results show that GT-GAN exhibits superior performance in the joint face super-resolution and frontalization task on the public face dataset Multi-PIE, which provides new ideas and references for this challenging topic.
Model-Agnostic Meta-Learning (MAML) and its variants have shown remarkable performance in scenarios characterized by a scarcity of labeled data during the training phase of machine learning models. Despite these succe...
Model-Agnostic Meta-Learning (MAML) and its variants have shown remarkable performance in scenarios characterized by a scarcity of labeled data during the training phase of machine learning models. Despite these successes, MAML-based approaches encounter significant challenges when there is a substantial discrepancy in the distribution of training and testing tasks, resulting in inefficient learning and limited generalization across domains. Inspired by classical proportional-integral-derivative (PID) control theory, this study introduces a Layer-Adaptive PID (LA-PID) Optimizer, a MAML-based optimizer that employs efficient parameter optimization methods to dynamically adjust task-specific PID control gains at each layer of the network, conducting a first-principles analysis of optimal convergence conditions. A series of experiments conducted on four standard benchmark datasets demonstrate the efficacy of the LA-PID optimizer, indicating that LA-PID achieves state-of-the-art performance in few-shot classification and cross-domain tasks, accomplishing these objectives with fewer training steps. Code is available on https://***/yuguopin/LA-PID. Copyright 2024 by the author(s)
Image steganography is the process of hiding secret information into an image while still keeping its original appearance intact. It plays a crucial role in secure communication and data storage. While deep learning-b...
详细信息
ISBN:
(数字)9798350379228
ISBN:
(纸本)9798350390780
Image steganography is the process of hiding secret information into an image while still keeping its original appearance intact. It plays a crucial role in secure communication and data storage. While deep learning-based techniques have shown notable progress in the field of image steganography, they still face challenges in extraction accuracy and robust security. This paper introduces an innovative image steganography method utilizing the diffusion model, which is the most powerful technique of image generation. Our method utilizes the sampling process of the diffusion model to hide the secret image in the generated image, and the receiver uses the secret key to extract the secret data accurately. The results of our experiments validate the effectiveness of our approach, outperforming existing methods in security, capacity, and extraction precision, which shows the potential for practical applications.
Accurate ultra-short-term photovoltaic (PV) power forecasting is crucial for the real-time scheduling of grid systems. However, the inherent variability of solar energy makes this task extremely challenging. To enhanc...
详细信息
This paper tackles the challenge of achieving prescribed-time synchronization (PTS) for stochastic complex networks (SCNs) under the framework of aperiodically intermittent control (AIC). The adoption of a time-varyin...
详细信息
ISBN:
(数字)9798350379228
ISBN:
(纸本)9798350390780
This paper tackles the challenge of achieving prescribed-time synchronization (PTS) for stochastic complex networks (SCNs) under the framework of aperiodically intermittent control (AIC). The adoption of a time-varying function ensures all states of the SCNs reach synchronization at the preset time. Compared with the PTS under the framework of deterministic systems, the introduction of stochastic disturbances has more practical significance. Introducing an auxiliary function makes the Lyapunov method mentioned in this paper valid under the AIC framework. Moreover, the AIC proposed in this paper relies on the average control rate instead of the infimum control rate, leading to less conservative results. At last, the viability of the theoretical results is confirmed by a simulation of a strongly connected digraph.
Converter-driven motor systems play crucial roles in contemporary industrial applications owning to the advantages of smooth starting and stepless speed regulation. For such fourth-order systems with both immeasurable...
详细信息
This paper studies a classical single pursuer and single evader pursuit-evasion *** pursuer attempts to capture the slower evader who aims to extend its lifetime during the *** simplify this question,requiring the eva...
详细信息
ISBN:
(数字)9789887581536
ISBN:
(纸本)9781665482561
This paper studies a classical single pursuer and single evader pursuit-evasion *** pursuer attempts to capture the slower evader who aims to extend its lifetime during the *** simplify this question,requiring the evader to take fixation strategy which is choosing the farthest point in its current dominant region as aimpoint and moving at a constant *** the pursuer is faster than the ***,the speed ratio is a *** instaneous state space will be partitioned into pursuer's dominant zone and evader's dominant zone by the generalized Apollonius *** pursuit strategy is based on minimizing the area of the evader's dominant ***,we propose a supermodular game for this ***,the existence of the Nash equilibrium is *** results based on Q-learning are presented to solve the problem,which shows the effectiveness of this method.
Identification of the type of combustion-supporting agents (CSAs) by an electronic nose (e-nose) is severely limited due to the absence of untested gas concentration in the e-nose training set. In order to solve this ...
详细信息
Recently,the formation control of multiple autonomous mobile robots(AMRs) have gained significant attention,and autonomous mobile robots(AMRs) have applied to all aspects of our ***-agent reinforcement learning is use...
详细信息
ISBN:
(数字)9789887581536
ISBN:
(纸本)9781665482561
Recently,the formation control of multiple autonomous mobile robots(AMRs) have gained significant attention,and autonomous mobile robots(AMRs) have applied to all aspects of our ***-agent reinforcement learning is used to solve the autonomously sequential decision-making problem of agents in a common environment with competition or ***,we present a utility function and a reward function to achieve formation control with collision avoidance for a rigid AMRs system,and build a simulation environment to meet environmental requirements based on MPE.
Automating complex industrial robots requires precise nonlinear control and efficient energy management. This paper introduces a data-driven nonlinear model predictive control (NMPC) framework to optimize control unde...
详细信息
暂无评论