search

yfxc / pseudo-3d-tensorflow

Tensorflow implement for Pseudo-3d-residual network.
21 stars 5 forks source link

Training problem #3

Open lihui52 opened 6 years ago

lihui52 commented 6 years ago

When I run train.py, I encountered such a problem： “ValueError: Cannot feed value of shape (10, 16, 160, 160, 3) for Tensor u'Placeholder:0', which has shape '(12, 16, 160, 160, 3)'” Could you help me, please?

yfxc commented 5 years ago

Sorry to see your problem so late,there may be a mismatch for the batch_size.

leesky1c commented 5 years ago

Modifying the batch_size will solve the problem.

AndyHon commented 5 years ago

When I run train.py, I encountered such a problem： “ValueError: Cannot feed value of shape (10, 16, 160, 160, 3) for Tensor u'Placeholder:0', which has shape '(12, 16, 160, 160, 3)'” Could you help me, please?

I also encountered the same problem, but the batchsize modification did not solve this problem. How did you solve it?Looking forward to your early reply.

yfxc commented 5 years ago

Please confire that you modificate the parameter firstly and then run the other codes.If you modificated the parameter after building the tensorflow graph,in this way,the modification would not take effect.

------------------ 原始邮件 ------------------ 发件人: "AndyHon"notifications@github.com; 发送时间: 2019年5月5日(星期天) 晚上10:13 收件人: "yfxc/pseudo-3d-tensorflow"pseudo-3d-tensorflow@noreply.github.com; 抄送: "忘却了蔚蓝"1512165940@qq.com; "Comment"comment@noreply.github.com; 主题: Re: [yfxc/pseudo-3d-tensorflow] Training problem (#3)

When I run train.py, I encountered such a problem： “ValueError: Cannot feed value of shape (10, 16, 160, 160, 3) for Tensor u'Placeholder:0', which has shape '(12, 16, 160, 160, 3)'” Could you help me, please?

I also encountered the same problem, but the batchsize modification did not solve this problem. How did you solve it?Looking forward to your early reply.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

AndyHon commented 5 years ago

Please confire that you modificate the parameter firstly and then run the other codes.If you modificated the parameter after building the tensorflow graph,in this way,the modification would not take effect. … ------------------ 原始邮件 ------------------ 发件人: "AndyHon"notifications@github.com; 发送时间: 2019年5月5日(星期天) 晚上10:13 收件人: "yfxc/pseudo-3d-tensorflow"pseudo-3d-tensorflow@noreply.github.com; 抄送: "忘却了蔚蓝"1512165940@qq.com; "Comment"comment@noreply.github.com; 主题: Re: [yfxc/pseudo-3d-tensorflow] Training problem (#3) When I run train.py, I encountered such a problem： “ValueError: Cannot feed value of shape (10, 16, 160, 160, 3) for Tensor u'Placeholder:0', which has shape '(12, 16, 160, 160, 3)'” Could you help me, please? I also encountered the same problem, but the batchsize modification did not solve this problem. How did you solve it?Looking forward to your early reply. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

Thank you.I have solved this problem, but when I was training, the loss remained as nan and accuracy remained as 0. I adopted data enhancement and added regularization items, but the problem has not been solved. Please tell me how I should deal with this problem.Thank you very much.

yfxc commented 5 years ago

As I know,chosing an inappropriate initializer or gradient explosion may lead to this problem.So you could chose other initializers or reduce your initial learning rate.Good luck!

------------------ 原始邮件 ------------------ 发件人: "AndyHon"notifications@github.com; 发送时间: 2019年5月6日(星期一) 上午10:27 收件人: "yfxc/pseudo-3d-tensorflow"pseudo-3d-tensorflow@noreply.github.com; 抄送: "忘却了蔚蓝"1512165940@qq.com; "Comment"comment@noreply.github.com; 主题: Re: [yfxc/pseudo-3d-tensorflow] Training problem (#3)

Please confire that you modificate the parameter firstly and then run the other codes.If you modificated the parameter after building the tensorflow graph,in this way,the modification would not take effect. … ------------------ 原始邮件 ------------------ 发件人: "AndyHon"notifications@github.com; 发送时间: 2019年5月5日(星期天) 晚上10:13 收件人: "yfxc/pseudo-3d-tensorflow"pseudo-3d-tensorflow@noreply.github.com; 抄送: "忘却了蔚蓝"1512165940@qq.com; "Comment"comment@noreply.github.com; 主题: Re: [yfxc/pseudo-3d-tensorflow] Training problem (#3) When I run train.py, I encountered such a problem： “ValueError: Cannot feed value of shape (10, 16, 160, 160, 3) for Tensor u'Placeholder:0', which has shape '(12, 16, 160, 160, 3)'” Could you help me, please? I also encountered the same problem, but the batchsize modification did not solve this problem. How did you solve it?Looking forward to your early reply. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

Thank you.I have solved this problem, but when I was training, the loss remained as nan and accuracy remained as 0. I adopted data enhancement and added regularization items, but the problem has not been solved. Please tell me how I should deal with this problem.Thank you very much.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

lccssr commented 5 years ago

Please confire that you modificate the parameter firstly and then run the other codes.If you modificated the parameter after building the tensorflow graph,in this way,the modification would not take effect. … ------------------ 原始邮件 ------------------ 发件人: "AndyHon"notifications@github.com; 发送时间: 2019年5月5日(星期天) 晚上10:13 收件人: "yfxc/pseudo-3d-tensorflow"pseudo-3d-tensorflow@noreply.github.com; 抄送: "忘却了蔚蓝"1512165940@qq.com; "Comment"comment@noreply.github.com; 主题: Re: [yfxc/pseudo-3d-tensorflow] Training problem (#3) When I run train.py, I encountered such a problem： “ValueError: Cannot feed value of shape (10, 16, 160, 160, 3) for Tensor u'Placeholder:0', which has shape '(12, 16, 160, 160, 3)'” Could you help me, please? I also encountered the same problem, but the batchsize modification did not solve this problem. How did you solve it?Looking forward to your early reply. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

Thank you.I have solved this problem, but when I was training, the loss remained as nan and accuracy remained as 0. I adopted data enhancement and added regularization items, but the problem has not been solved. Please tell me how I should deal with this problem.Thank you very much.

have u solve this problem? I meet the same problem with u.Please tell me how I should deal with this problem.Thanks