可以使用扩散模型来降低图像吗？
我对扩散模型有些熟悉，因为它们是学习图像的一些潜在变量表示的过程。这是通过学习如何消除渐进高斯噪声的影响来完成的。我想知道我是否可以使用扩散模型来代诺图像？有人做过吗？
I'm somewhat familiar with diffusion models, in that they're a process to learn some latent variable representation of images. This is done by learning how to undo the effect of progressive Gaussian noise. I am wondering if I could use diffusion models to denoise images? Has anyone done this before?
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（2）
是的，他们可以并且有关于
我非常熟悉言语denoising。对于基于改进的脱氧扩散概率模型
模型使用扩散模型来降解的无限潜力，因为它们能够灵活地对任何分布的分布建模和可以分析评估的可拖动模型的制定。
Yes, they can and there are research publications on that Denoising Diffusion Probabilistic Models.
I am very conversant with speech denoising. For the speech denoising based on the improved diffusion probabilistic models presented in Improved Denoising Diffusion Probabilistic Models
I believe the are unlimited potentials to the use of diffusion models for denoising because of their ability to flexibly model any distribution and the formulation of tractable model that can be evaluated analytically.
剥落图像就像图像的线性反问题。在使用扩散模型的各种测量噪声下，在各种量的测量噪声下进行了多项努力来解决相关问题，例如超分辨率，脱毛，涂料和着色。
研究工作DDRM-降级扩散恢复模型也相同。在重建质量，感知质量和运行时，DDRM的表现优于当前领先的无监督方法。这是读取 https://arxiv.org/abs/2201.11793
Denoising images is like a linear inverse problem for images. There have been multiple efforts to solve the related problems such as super-resolution, deblurring, inpainting, and colorization under various amounts of measurement noise using diffusion models.
Research work DDRM - Denoising diffusion Restoration model tackles the same. DDRM outperforms the current leading unsupervised methods on the diverse ImageNet dataset in reconstruction quality, perceptual quality, and runtime. Here is the read https://arxiv.org/abs/2201.11793