Training mnist_compiled_data get error

[Hai le hong]

Hello,

I have a newbie question, in demos/mnist/mnist.conf file, i change

ubyte files by .mat files frommnist_compiled_data.zip file (http://eblearn.sourceforge.net/beginner_tutorial2_train.html)

i get following error when i train with the modified config file above:

"warning: convolution connection table is not using all inputs in layer conv0 the maximum input index used by the table is 0 but the input is 3x36x36

Error: 10x2x2 and 10x1x1 should have the same number of elements, in ebl::idx_sqrdist at e:\research\eblearn\trunk\core\libidx\include\idxops.hpp:1803"

Thanks for your help.

[rouuuge]

i got the same error.

i have build a new data set with dscompile like : 

dscompile <path> -outdir <path> -dname test_yellow -dims 64x64x3,32x32x3,16x16x3 -kernelsz 9x9,9x9,9x9 -channels YUV -colornorm_across 1 -lpyramid 3 -disp

now i get the error: Error: 3x32x32 and 3x64x64 should have the same number of elements

i use the covnet config from the mnist-tutorial.

thanks

[Zizhou Liu]

Hi le hong,

I got the save error.

Have you solved this?

Cheers,

Zizhou

[zhza...@gmail.com]

Hello,
Have you solve this problem? I met the same error.


Hui
在 2013年4月10日星期三UTC+8下午12时26分42秒，Hai le hong写道：

[Manuel Roth]

Nope

—
Von Mailbox gesendet

--
You received this message because you are subscribed to a topic in the Google Groups "eblearn" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/eblearn/8ZcQEHSJty4/unsubscribe.
To unsubscribe from this group and all its topics, send an email to eblearn+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[manju...@gmail.com]

Hey,
I know this is like a year after this query was posted. But I got the same error right now and I resolved it. So this for future reference just in case anyone bumps into the same problem later on. This error pops up because finally you should converge to 10 values - 1 row, 1 column. But due to some internal problem, you are ending up in 2 rows, 2 cols with 10 values. When I checked my log, in the very first convolution, the 32x32 was supposed to give me a 28x28 feature map when convolved with a 5x5 filter but instead it gave a 36x36. This caused an extra row to be added in the result. This happened because z_pad was set to 5x5. This padding caused the convolution step to give the unexpected output. So I changed the 5x5 at the padding to something less than 5x5 - i made it 1x1 and this resolved the error.
Analyze your log and see whats going wrong. It helps! :)

[wangz...@gmail.com]

在 2014年11月29日星期六 UTC+8上午3:00:25，manju...@gmail.com写道：

> Hey,
> I know this is like a year after this query was posted. But I got the same error right now and I resolved it. So this for future reference just in case anyone bumps into the same problem later on. This error pops up because finally you should converge to 10 values - 1 row, 1 column. But due to some internal problem, you are ending up in 2 rows, 2 cols with 10 values. When I checked my log, in the very first convolution, the 32x32 was supposed to give me a 28x28 feature map when convolved with a 5x5 filter but instead it gave a 36x36. This caused an extra row to be added in the result. This happened because z_pad was set to 5x5. This padding caused the convolution step to give the unexpected output. So I changed the 5x5 at the padding to something less than 5x5 - i made it 1x1 and this resolved the error.
> Analyze your log and see whats going wrong. It helps! :)

Thank you very much. It's very helpful!!!!!! ^_^