fcn.pytorch icon indicating copy to clipboard operation
fcn.pytorch copied to clipboard

Published 20 hours ago • verified publisher icon L1aoXingyu

Environment configuration

Open chaowentao opened this issue 6 years ago • 12 comments

Could you please provide detailed environment configuration description and completed training process description (including data sets download, etc.)? Thanks a lot~

chaowentao avatar Dec 05 '18 12:12 chaowentao

I am sorry that I forgot about the dataset parts. I will add this as soon as possible. Btw, for environment configuration, you only need to install the package I list in requirements.

L1aoXingyu avatar Dec 06 '18 06:12 L1aoXingyu

Thanks a lot. I am your students from cs231n camp. Thanks again.

chaowentao avatar Dec 06 '18 06:12 chaowentao

I have added the dataset configuration. Hope this can help.

L1aoXingyu avatar Dec 07 '18 05:12 L1aoXingyu

I reconfigured the environment again, the result is still the original problem. I guess it may be a problem with the windows system, then I will try it under the Ubuntu system. Thank you. _20181207143454

chaowentao avatar Dec 07 '18 06:12 chaowentao

you can try to set num_workers=0 in DataLoader

L1aoXingyu avatar Dec 19 '18 08:12 L1aoXingyu

Can you run this training script? If you have any other problems, you can tell me in this issue and in case others meet the same problems, too.

L1aoXingyu avatar Dec 29 '18 06:12 L1aoXingyu

Thanks.i meet some problem.and it work now.Thanks.

Merofine avatar Mar 17 '19 07:03 Merofine

but it is too slow. 捕获 @L1aoXingyu

Merofine avatar Mar 17 '19 07:03 Merofine

It is very slow because the batch size is 1, you can set the batch size = 32, then do some preprocessing, such as center crop etc, to make the input image size same.

L1aoXingyu avatar Mar 17 '19 07:03 L1aoXingyu

Yes, I changed it to 32, but it still reported a mistake. It is estimated that there is no pretreatment. Actually, I'm particularly interested in your project, but I'm just a whitewash. Could you please give me more detailed steps? 捕获

Merofine avatar Mar 18 '19 02:03 Merofine

Yes, I changed it to 32, but it still reported a mistake. It is estimated that there is no pretreatment. Actually, I'm particularly interested in your project, but I'm just a whitewash. Could you please give me more detailed steps? 捕获 Hi, I've meet the same problem. Have you fixed it? Thanks a lot.

KinsoZHENG avatar Dec 28 '19 22:12 KinsoZHENG

Yes, I changed it to 32, but it still reported a mistake. It is estimated that there is no pretreatment. Actually, I'm particularly interested in your project, but I'm just a whitewash. Could you please give me more detailed steps? 捕获 Hi, I've meet the same problem. Have you fixed it? Thanks a lot.

I have tried to fix the batch size problem. After changing the batch size in config/defaults.py, please add
img = img.resize((x, x)) # the size what you wanna after https://github.com/L1aoXingyu/fcn.pytorch/blob/7f592407f41325375baff5b3514567fb4c5f9a62/data/datasets/voc.py#L34 label = label.resize((x, x)) # the same size of before after https://github.com/L1aoXingyu/fcn.pytorch/blob/7f592407f41325375baff5b3514567fb4c5f9a62/data/datasets/voc.py#L36 It can make the program work.

KinsoZHENG avatar Dec 29 '19 21:12 KinsoZHENG