search

jakeret / tf_unet

Generic U-Net Tensorflow implementation for image segmentation
GNU General Public License v3.0
1.9k stars 748 forks source link

conflict between tensorboard and tensorflow when they are running at the same time #232

Closed GonzaloMoreno closed 5 years ago

GonzaloMoreno commented 5 years ago

Hello. I am running a tensorflow code and I want to see how the accuracy of the model changes while the training is running. However, When I activate tensorboard, the proccess of training in tesorflow stops and generate this error:

OP_REQUIRES failed at save_restore_v2_ops.cc:137

Failed to rename: ./unet_trained\model.ckpt.index, ......Access Denied.

This error only happens when tensorflow is running at the same time that I open the ip http://127.0.0.1:6006/. I have this problem with windows 10, tensorflow 1.11.0 and tensorbard 1.11.0

How could I solve this?

jakeret commented 5 years ago

I haven't used to code on windows - I suspect this might origin because of the OS and not due to the tf_unet implementation. Might be worth checking if other tensorflow users have expirenced the same issue on windows

GonzaloMoreno commented 5 years ago

Hi: Thanks for your fast answer. I am new working in tensorflow and your tf_unet. I see that this is happening to other Windows users. However, I see that this happens because it writes the file in the same log folder and the class does not have a parameter to save the file in other different path.

GonzaloMoreno commented 5 years ago

Hi: I changed the line 405 of the file unet.py to save the model in a folder called "model_ckptf" instead of the default "output_path". Two different path solves the conflict and now it works for linux either Windows.