search

fenglinglwb / PSM

Image Inpainting via Iteratively Decoupled Probabilistic Modeling
67 stars 0 forks source link

网络结构 #3

Open sunohm opened 1 year ago

sunohm commented 1 year ago

你好，这篇论文模型没不太明白，想请教一下，网络整体结构是采用编码器解码器结构，解码器使用的styleGAN，那编码器是是卷积块堆叠起来的，还是使用的unet，希望得到你的回复。

fenglinglwb commented 1 year ago

Thanks for your interest. We use stacked convolutional blocks to build the encoder.

sunohm commented 1 year ago

感谢你的回复，我看了补充材料里面体系结构详细信息，图四U-Net architecture 是整体结构吗

light&￡ @.***

 

------------------ 原始邮件 ------------------ 发件人: "fenglinglwb/SDM" @.>; 发送时间: 2023年2月28日(星期二) 上午9:34 @.>; @.**@.>; 主题: Re: [fenglinglwb/SDM] 网络结构 (Issue #3)

Thanks for your interest. We use stacked convolutional blocks to build the encoder.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

fenglinglwb commented 1 year ago

Exactly.

sunohm commented 1 year ago

好的，谢谢，我还是有点迷糊，通过整体结构看，解码器是使用的StyleGAN中的D吗？

light&￡ @.***

 

------------------ 原始邮件 ------------------ 发件人: "fenglinglwb/SDM" @.>; 发送时间: 2023年2月28日(星期二) 上午9:50 @.>; @.**@.>; 主题: Re: [fenglinglwb/SDM] 网络结构 (Issue #3)

Exactly.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

fenglinglwb commented 1 year ago

Yep. We use StyleGAN 2 as the decoder.

sunohm commented 1 year ago

对于文章，编码器卷积堆叠而成是噪声扩散过程，解码器是去除噪声的过程是这样吗？那是使用StyleGAN2的生成器作为encoder-decoder架构的解码器吗？

light&￡ @.***

 

------------------ 原始邮件 ------------------ 发件人: "fenglinglwb/SDM" @.>; 发送时间: 2023年2月28日(星期二) 上午10:17 @.>; @.**@.>; 主题: Re: [fenglinglwb/SDM] 网络结构 (Issue #3)

Yep. We use StyleGAN 2 as the decoder.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

fenglinglwb commented 1 year ago

Our model is NOT a denoising diffusion model. Diffusion-based methods in traditional inpainting work refer to gradually propagating useful information from undamaged regions to missing regions. We follow this line and name our model a spatial diffusion network.

sunohm commented 1 year ago

好的，非常感谢。

light&￡ @.***

 

------------------ 原始邮件 ------------------ 发件人: "fenglinglwb/SDM" @.>; 发送时间: 2023年2月28日(星期二) 中午11:34 @.>; @.**@.>; 主题: Re: [fenglinglwb/SDM] 网络结构 (Issue #3)

Our model is NOT a denoising diffusion model. Diffusion-based methods in traditional inpainting work refer to gradually propagating useful information from undamaged regions to missing regions. We follow this line and name our model a spatial diffusion network.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>