search

leeyeehoo / CSRNet-pytorch

CSRNet: Dilated Convolutional Neural Networks for Understanding the Highly Congested Scenes
645 stars 258 forks source link

关于patch的size问题 #2

Open ShwanMario opened 6 years ago

ShwanMario commented 6 years ago

您好,感谢您百忙之中抽空解答我的疑问.

我有一个问题想请教一下,就是你们在crop 1/4 size的原图的patch后,是直接用patch拿来训练还是resize patch到其他的分辨率后再训练呢?

感谢您的赐教!

leeyeehoo commented 6 years ago

直接拿patch做训练

markshih91 commented 6 years ago

您好,感谢您百忙之中抽空解答我的疑问.我想请教您如下问题：

1、输入图像大小不同，训练时，batch_size只能设置成1吗？ 2、或者在这种情况下，怎么时batch_size大于1？

谢谢！！！

leeyeehoo commented 6 years ago

只能设成1,或者你自己改一下image.py文件,比如说crop一个fixed的大小的输入.

zhengmiao1 commented 4 years ago

你好，您的crop好像没有使用呀？因为train.py文件中train=True，而在imge.py中进行crop需要是train在False的情况下

ConstantSun commented 4 years ago

Hello, your crop does not seem to be used? Because train=True in the train.py file, and crop in imge.py needs to be train in the case of False

Yes, I totally agree with @lingkongyingsun . @leeyeehoo your crop does not use, in image.py file , you wrote: if False: ....( which is always not to be executed ) . According to the paper, we need to augment data before training, you did miss that , right @leeyeehoo ??