How to generate customized convolutional filters

[Hoang Nguyen]

Hello,

I would like to generate convolutional filters (for 2nd, 3rd, ... stages) corresponding to my dataset, but not randomly. My questions are:

1/ What are the default convolutional filters in EBLearn? Are they generated randomly?

2/ Is there any component in EBLearn generating custom convolutional filters (which may be 'conv_table' parameters in configuration files)?

Please help me on this.

Regards,

Hoang.

[the_minion]

Hey,

All the convolution filters in eblearn's convolution layers are learnt (i.e.. customized).

They are initialized randomly, and during the training phase, they're learnt.

[Hoang Nguyen]

Thanks the_minion. I'm so appreciate your response.

According to http://yann.lecun.com/exdb/publis/pdf/jarrett-iccv-09.pdf, using trained filters (i.e., by Predictive Sparse Decomposition) results in better recognition in comparison with random ones. Do you have any idea about using PSD in EBLearn?

Regards.

--
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/bvcxfONAseM/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/groups/opt_out.

[the_minion]

PSD, in eblearn, you'd have to write a lot of code.

However, if you use Torch-7, koraykv (one of the authors of the paper) has the code available in https://github.com/koraykv/unsup/

[Hoang Nguyen]

Thanks so much. This is cool. I'll check it out.

Best regards,

Hoang.

[Hoang Nguyen]

Hi the_minion,

Unsup works just fine to me now. But I wonder how to apply it into EBLearn. I mean I don't know how to produce .bin (and .ascii) input file of Unsup from images. Would you please show me how to combine them two?

I also found Lush version of PSD of Koraykv. I tried using output of dscompile as its input, but it doesn't work.

Regards.