Apply GANs week2

Image-to-Image Translation

What is paired translation?
When you have corresponding input-output images that you can gather training data for.

Correct! Paired translation means that you have input-output pairs that map exactly onto each other (1-to-1).

PatchGAN

What is the output of the Pix2Pix discriminator (PatchGAN)?
A matrix with values between 0 and 1 indicating realness.

Pix2Pix: U-Net

在这里插入图片描述

  1. Pix2Pix先用U-net来做image segmentation, 再用U-net和segmentation后的image来做generate.

  2. Skip connection可以改善back propagation的性能,在propagation中也可以提供encode时损失的信息

  3. 在decoder前几个block中用了dropout,来使block在不同的Iteration中学到不同的东西, inference和test时关掉dropout(但要do scale by probability)

What is the purpose of the skip connections in the Pix2Pix generator?
They allow information to flow to later layers by reducing the vanishing gradient problem, which occurs when there are too many layers.

Correct! Skip connections help prevent the vanishing gradient problem when gradients get too small when there are many layers because they
are weighed less and less as the layers get deeper.

Pix2Pix: Pixel Distance Loss Term

Why is pixel distance used for L1 regularization in Pix2Pix?
Because the real output image and the generated image should be encouraged to be similar.

Correct! Since they are so correlated, you can use the absolute difference or pixel difference.

在这里插入图片描述

Put together

在这里插入图片描述

在这里插入图片描述
在这里插入图片描述

  • 0
    点赞
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值