Demo fails (CUDNN_STATUS_SUCCESS (8 vs. 0)) about deep-image-analogy HOT 9 CLOSED

Actually the demo was built with cudnn, so cudnn is needed to run it. But you can recompile it without cudnn to reduce some memories.

from deep-image-analogy.

I have cuDNN extracted, though. Have I installed it in the wrong place or something?

from deep-image-analogy.

Hello. I met the same problem. Have you solved this problem yet?

from deep-image-analogy.

Have you tried to change the value of ratio from 0.5 to 0.2 which could reduce some memories.

from deep-image-analogy.

Changing the ratio worked, however i still got some errors.
Finding nearest neighbor field using PatchMatch Algorithm at layer:conv1_1. Saving flow result. Finished finding ann. Time : 21.577 WARNING: Logging before InitGoogleLogging() is written to STDERR F0918 15:20:57.037252 3020 syncedmem.hpp:31] Check failed: error == cudaSuccess (4 vs. 0) unspecified launch failure *** Check failure stack trace: ***

Lowering ratio to 0.1 worked flawlessly, however the results...
There are no visible changes at ratio this small! xd

from deep-image-analogy.

Try to disable the time limit (TDR). #25

from deep-image-analogy.

hello, @rozentill . I try to run the demo but I meet this error too. I don`t know how to disable the time limit. If you could tell me how to disable the time limit,I will appreciate you very much.

from deep-image-analogy.

You've probably figured out yourself, but anyways.. It kind of worked yet, my first results weren't as astonishing, but the example was rather suitable for Neural Style Transfer.
Would be noice if the Stereoscopic Style Transfer would get a repository with binaries too!

https://www.pugetsystems.com/labs/hpc/Working-around-TDR-in-Windows-for-a-better-GPU-computing-experience-777/
https://forums.geforce.com/default/topic/503962/tdr-fix-here-for-nvidia-driver-crashing-randomly-in-firefox/
Change these three values in regedit.exe to a high value (i used 60000 from about 3-7)

Also, the piece of software has some problems here in there. Rather a minor is that when specifing the models path, the code automatically adds "vgg19/" to it..
It hangs for a while on the Reading dangerously large protocol message, too. I tried to copy the 576MB weights file to a simulated RAM disc, and it didn't quickened it much either. Ya could implement batch processing that could save some of the precious time wasted on loading it..

from deep-image-analogy.

@ad48hp Thank you!!! Although I have solve this problem. lol

from deep-image-analogy.