How to solve this problem when the program is running?

[HTaCheng]

[nmerrill67]

I have had trouble with CUDA10.0 and TensorFlow recently. You should try CUDA10.1, or try compiling TensorFlow from source.
https://www.tensorflow.org/install/gpu

[HTaCheng]

Thank you very much for your suggestions. I tried CUDA 10.1 before, but there are still problems. Can you tell me the CUDA, CUDNN, tensorflow and Python versions when you ran this program?

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "rpng/calc2.0" <[email protected]&gt;; 发送时间:&nbsp;2020年11月20日(星期五) 凌晨2:06 收件人:&nbsp;"rpng/calc2.0"<[email protected]&gt;; 抄送:&nbsp;"他城"<[email protected]&gt;;"Author"<[email protected]&gt;; 主题:&nbsp;Re: [rpng/calc2.0] How to solve this problem when the program is running? (#11) I have had trouble with CUDA10.0 and TensorFlow recently. You should try CUDA10.1, or try compiling TensorFlow from source. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[nmerrill67]

They are outdated at this point. I think it was on a CUDA 9.X. Instead, you can try the exact CUDA, CUDNN, and TensorFlow shown in the link I sent. I suspect it's a CUDNN compatability issue. Ensure you have CUDA10.1 and CUDNN 7.6 as it says. With that, you should be able to use the current TensorFlow from pip.

[HTaCheng]

Thank you very much for your answers, I finally solved this problem according to your tips. But there are a lot of results like this during training, is there a problem with my training process?

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "rpng/calc2.0" <[email protected]&gt;; 发送时间:&nbsp;2020年11月21日(星期六) 凌晨1:05 收件人:&nbsp;"rpng/calc2.0"<[email protected]&gt;; 抄送:&nbsp;"他城"<[email protected]&gt;;"Author"<[email protected]&gt;; 主题:&nbsp;Re: [rpng/calc2.0] How to solve this problem when the program is running? (#11) They are outdated at this point. I think it was on a CUDA 9.X. Instead, you can try the exact CUDA, CUDNN, and TensorFlow shown in the link I sent. I suspect it's a CUDNN compatability issue. Ensure you have CUDA10.1 and CUDNN 7.6 as it says. With that, you should be able to use the current TensorFlow from pip. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[nmerrill67]

I'm not sure what the issue is. Can you specify? TensorFlow warnings?

[HTaCheng]

Tensorflow did not report errors or warnings. Sorry, as a beginner, some questions may seem a bit stupid.&nbsp; My question and understanding is this.During the entire training process, each input picture will generate an image after passing through the network. For the entire training set, I think the output image corresponding to each input image should be similar to the second image below. But most of my output pictures are blank like the first one. Is this phenomenon reasonable? Can you help me explain the problems in my training process?

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "rpng/calc2.0" <[email protected]&gt;; 发送时间:&nbsp;2020年11月22日(星期天) 上午7:08 收件人:&nbsp;"rpng/calc2.0"<[email protected]&gt;; 抄送:&nbsp;"他城"<[email protected]&gt;;"Author"<[email protected]&gt;; 主题:&nbsp;Re: [rpng/calc2.0] How to solve this problem when the program is running? (#11) I'm not sure what the issue is. Can you specify? TensorFlow warnings? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.