Imageai: Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR

[bestbugwriter]

I use docker to run example, and I find a bug,

I find this link to resolve the problem, but how to set tensorflow gpu operation in imagiai?
https://github.com/tensorflow/tensorflow/issues/24496

[rola93]

how to set tensorflow gpu operation in imagiai?

It is already set by default.

Are you sure that your tensorflow instalation/configuration is ok?

[bestbugwriter]

I was exec in docker, the images: nwpu/imageai-gpu:local
the Dockerfile : https://github.com/yywf/dev-env/blob/local/imageai-gpu.df

[rola93]

You should look at this issue

[HairyFotr]

Every solution in that thread requires the modification of tf session (which is controlled by ImageAI), but it's also possible to do it with an envrionmental variable: TF_FORCE_GPU_ALLOW_GROWTH=true

[ajsirsik]

Hey there

Hope you'all doing well :)

I am also getting same error message, even after setting GPU_ALLOW_GROWTH=true and process_gpu_memory_fraction limit, issue persists. May I know any other way to get rid of this issue, and use gpu to run imageai code.

Thanks in advance

Best regards