Model train performance

[guanfuchen]

The code is great, and I use the code to get coarse result, but I find it difficult to get the best performance on CamVid or other dataset, like segnet, fcn-8s performance in the origin paper. Does anyone get a good result using the code and share the trick?

[johnathanchiu]

Does your image turn out completely black? I'm not sure why but mine does.

[guanfuchen]

@johnathanchiu I can train it using the old version in the pytorch-semseg project, what is the log?

[johnathanchiu]

what model are you using? I am using the icnet one and it says its pretrained but it doesn't work.

[jizongFox]

@johnathanchiu I can train it using the old version in the pytorch-semseg project, what is the log?

Hi, I saw this is visdom, what is the old version that works?

[SSyangguang]

what model are you using? I am using the icnet one and it says its pretrained but it doesn't work.

Hi, have you solved the problem? My images are also completely black.

[xdsonglinliu]

@SSyangguang I have met the same problem. have you solved it?