影子GAN-前:重权自我注意力基于面具的影子去除
Jianyi Hu1, Shuhuan Wen1, Jiaqi Li1
1Engineering Research Center of the Ministry of Education for Intelligent Control System and Intelligent Equipment, Yanshan University, Qinhuangdao, China; Key Laboratory of Industrial Computer Control Engineering of Hebei Province, Yanshan University, Qinhuangdao, China; Key Lab of Intelligent Rehabilitation and Neuroregulation of Hebei Province, Yanshan University, Qinhuangdao, China.
影子GAN-Former通过使用非影子区域进行指导来有效地去除影子. 这种新的变压器和生成对抗网络 (GAN) 模型可以提高图像质量和细节恢复.
科学领域:
- 计算机视觉 计算机视觉
- 人工智能的人工智能
- 图像处理 图像处理
背景情况:
- 影子去除对于图像恢复至关重要,但由于现有方法的不一致性,具有挑战性.
- 许多方法忽略了来自非影子区域的有价值信息,影响了图像质量.
研究的目的:
- 开发一种高效的混合模型,以改善影子清除.
- 为了利用非影子区域信息,实现更一致,更准确的无影子图像重建.
主要方法:
- 拟议的ShadowGAN-Former,是一种混合变压器和生成对抗网络 (GAN).
- 引入了多头转移注意力 (MHTA) 和门式前网络 (Gated FFN) 以实现高效的特征提取.
- 开发了影子注意力重权模块 (SARM),以根据影子与非影子区域相关性重权注意力地图.
主要成果:
- 在ISTD和SRD数据集上,ShadowGAN-Former在最先进的方法上表现出优越的性能.
- SARM模块显著提高了5.42%的峰值信号噪声比 (PSNR).
- SARM模块减少了14.76%的根平均平方误差 (RMSE).
结论:
- 影子GAN-Former通过结合上下文信息提供了一种有效的解决方案来消除影子.
- 提出的注意力机制和模块增强了模型准确恢复阴影影响的图像区域的能力.
更多相关视频
13:00Measuring Attention and Visual Processing Speed by Model-based Analysis of Temporal-order Judgments
Published on: January 23, 2017
06:46Investigating the Deployment of Visual Attention Before Accurate and Averaging Saccades via Eye Tracking and Assessment of Visual Sensitivity
Published on: March 18, 2019
相关概念视频
Self-Presentation: Self-Monitoring and Self-Handicapping
Weighted Mean
For example, consider the number of goals scored in the matches of a tournament. While computing the average number of goals scored in the tournament, it may be more important to...
Masking and Demasking Agents
There are many masking agents, such as cyanide, fluoride, triethanolamine, thiourea, and 2,3-bis(sulfanyl)propan-1-ol (formerly 2,3-dimercapto-1-propanol), with the masking agent chosen based on...
Apparent Weight
Consider a person standing on a bathroom scale inside an elevator. If the scale is accurate at rest, its reading equals the...
Impression Management Techniques IV: Altercasting
