ValueError: Variable d_h0_conv/w/Adam/ does not exist

[chihyaoma]

When running on ImageNet, I got this error.

ValueError: Variable d_h0_conv/w/Adam/ does not exist, or was not created with tf.get_variable(). Did you mean to set reuse=None in VarScope?

After searching online for a while, I found that this issue is introduced by using TensorFlow 0.12.

While using tf.variable_scope() solve the issue on DCGAN, I found it difficult to apply the same techniques on improved-gan.

If anyone faced the same issue, can you please share with me how you solve it?

[zhangqianhui]

I also got this error by running DCGAN.I think it is caused by tensorflow 0.12.
So , I wrote the DCGAN code by myself , it don't has this problem.

https://github.com/zhangqianhui/Conditional-Gans

hoping it can help you.

[chihyaoma]

@zhangqianhui I followed the commit from your repository, but unfortunately, it didn't solve this problem for improved-gan. I believe it's because the logic and code structure in improved-gan is still different from DCGAN.

I will try again sometime soon and get back to you if I solve it.

[zhangqianhui]

@chihyaoma It's not caused by this commit. But in tensorflow 0.11 , the project is ok

[zhangqianhui]

@chihyaoma Have you solved it?

[chihyaoma]

@zhangqianhui I haven't had the time to work on it. I will make sure to update this post if I have the time to fix the problem. Thanks for asking.

[zhangqianhui]

@hans @jonasschneider @welinder @yaroslavvb @goodfeli
have you test the project in tensorflow 0.12?
thanks for your answer

[goodfeli]

I haven't tested in in 0.12.

[zhangqianhui]

@goodfeli the error "d_h0_conv/w/Adam/" confused me.
And DCGAN also have this problem , but it have updated ans solved it.
Hope you test it and update.Thanks.

[goodfeli]

A PR is welcome but I'm not likely to do much maintenance on this repo. It's research code released for the purpose of openness and reproducibility, not a library that I'm committing to support indefinitely. It's not feasible for me to personally provide ongoing support for all previously released research code, especially since that body of code grows over time.

…

On Fri, Dec 30, 2016 at 6:56 PM, jichao zhang ***@***.***> wrote: @goodfeli <https://github.com/goodfeli> the error "d_h0_conv/w/Adam/" confused me. And DCGAN also have this problem , but it have updated ans solved it. Hope you test it and update.Thanks. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#12 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAXrGq8lFBNy7vC89FuwJ7RzkDx-mMSZks5rNcRogaJpZM4LTZTs> .

[zhangqianhui]

@goodfeli ok , If I have solved it , I will make a PR

[zhangqianhui]

I have solved this problem , and I will test it , and make a PR @goodfeli @chihyaoma

[zhangqianhui]

@chihyaoma I have updated codes
https://github.com/zhangqianhui/improved-gan
You can test it.Maybe it's have some problems.

[goodfeli]

From a quick glance, it looks basically OK, but two small issues: - Would it still support older versions of tensorflow, or only 0.12? - It looks like it changes the hard-coded number of GPUs from 4 to 1. We'd need to leave it at 4 if we were to merge the code as is, since that's how the experiments in the paper were conducted.

…

On Sun, Jan 1, 2017 at 7:28 PM, jichao zhang ***@***.***> wrote: @chihyaoma <https://github.com/chihyaoma> I have updated codes https://github.com/zhangqianhui/improved-gan You can test it.Maybe it's have some problems. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#12 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAXrGkGlQsbQF4C_14G38quH8vppUP_wks5rOG7QgaJpZM4LTZTs> .

[zhangqianhui]

@goodfeli ,Just supported tf0.12. the code need more changes , and I will test and updated to support 4 gpu.

[zhangqianhui]

@goodfeli I update the code again .

https://github.com/zhangqianhui/improved-gan

[brb-chen]

@zhangqianhui
did u got any contact info, such as email ?
how can i contact with u ?

[goodfeli]

I get the github notifications, so you can contact me that way.
I don't work at OpenAI anymore, so maybe we need to have @TimSalimans approve your PR when you're ready to make it. I can still review PRs that relate to the ImageNet code.