Sheng You

@primejava Maybe there is a problem of the inputting image size~

@yzceig Make sure your typing language has been shifted to ENG. Press `n` only when the activated window is `OpenCV` instead of `CMD`. Or u can add some debug messages...

@yzceig I'm afraid so. Use pictures with smaller size to try again and I strongly recommend you to run with GPU~

@yzceig The cuda toolkit is installed firstly whether 9.0 or 10.0 is supported by `PyTorch`. Be careful you should use the correct command line to install `PyTorch` which is written...

I have no idea, it looks like the weights file is not matched with the structure. PLZ make sure u have load the correct weight file.

你可以再debug一下，图像尺寸应该是在预处理的时候根据yml里的配置resize的。

@meansw 查了下应该是python版本问题，建议换python3.6或3.5.4，之后重新安装对应版本的pytorch

@xmjiayou There must be something wrong, transfer learning is quite fast that u can see the results obviously better in just one day!

It could work but it is really hard to restore background of such large area.

coursera上可以直接做编程题的，选择旁听不收费