Structure-emphasized-Multimodal-Style-Transfer icon indicating copy to clipboard operation
Structure-emphasized-Multimodal-Style-Transfer copied to clipboard

Published 20 hours ago • verified publisher icon irasin

result is black

Open vbqgyy opened this issue 3 years ago • 8 comments

(PY388) G:\proj\YUs_Ai\02PYTORCH\00style_transfer\HD\00Structure-emphasized-Multimodal-Style-Transfer-master\SEMST_Original>python test.py -c ../content/avril.jpg -s ../style/horse.jpg CUDA available: GeForce RTX 3090 SEMST_Original_model_state.pth loaded result saved into files starting with avril_horse

———————————————————————————————————— All is good, but the result is black win10 pytorch 1.8.1 python 3.8 rtx 3090 cuda 11.1 avril_horse_style_transfer_demo

vbqgyy avatar Apr 30 '21 05:04 vbqgyy

Does any error occur?

I have not tested on the newest Pytorch yet. Show me the log if possible.

irasin avatar May 01 '21 06:05 irasin

No. I tested it on the cpu and everything is good.Where is the log file?

Sin @.***> 于2021年5月1日周六 下午2:00写道：

Does any error occur?

I have not tested on the newest Pytorch yet. Show me the log if possible.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/irasin/Structure-emphasized-Multimodal-Style-Transfer/issues/2#issuecomment-830557817, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFKY26SI5EDHZKVQXDL4JDTTLOKIHANCNFSM433QQSDQ .

vbqgyy avatar May 02 '21 10:05 vbqgyy

Hi, @vbqgyy

I tested it on PyTorch1.8 cpu and it seems OK. Check your env again please.

avril_horse_style_transfer_demo

image

irasin avatar May 05 '21 08:05 irasin

There is no problem with cpu, only gpu has a problem。

vbqgyy avatar May 06 '21 00:05 vbqgyy

@vbqgyy

Sorry, I will check it later

irasin avatar May 06 '21 03:05 irasin

That would be great. Thank you.

vbqgyy avatar May 06 '21 06:05 vbqgyy

Hi, @vbqgyy

I have tested on cuda11 + Pytorch1.8,1 and the result is wrong indeed. And the reason why the result is black, is because the output of Model is Nan.

However, when I tested it on cuda10.1 + Pytorch1.7.1, the result is OK again.

I guess there are some version compatibility problems here. I will fix it if I have time. Or you can check it by yourself. Best wish.

irasin avatar May 06 '21 10:05 irasin

@irasin Thank you for taking the time to do the test.I am just a beginner.I guess it will take a lot of time and luck to complete this thing.

vbqgyy avatar May 07 '21 01:05 vbqgyy