Caffe features to libsvm

[Jos van de Wolfshaar]

Hi everybody,

While I was looking on the web for scripts to convert Caffe features to libsvm compatible text files I could not find anything useful. Hence, I decided to write my own script that extracts the features in the right format. The script can be found at:

https://github.com/jostosh/caffe-utils

It can also be used to extract everything to a plain text format. I hope some of you find it useful!

For downloading libsvm check out:

http://www.csie.ntu.edu.tw/~cjlin/libsvm/

Cheers,

Jos

[Luca Minciullo]

Hi,

first, thank you for you contribution.

What do you exactly mean by mini-batches in your algorithm?

Best,

Luca

[Jos van de Wolfshaar]

Hi,

Good to hear that you like it.

I have to admit the 'mini' in mini-batch is redundant and confusing. It's the number of batches that are 

processed. It's more or less a legacy from another script that I used to write this. 

Anyway, let's say that you chose to use 100 mini-batches of 128 images each. That would result in a

text file that contains the features of 100 * 128 images in total.

Hope this makes things clear.

Cheers,

Jos

Op maandag 21 september 2015 14:42:03 UTC+2 schreef Luca Minciullo:

[Qingkai Lu]

Hi Jos,

Thank you for sharing your script. I am new to caffe and this might be exactly what I need. 

I have one question about how to use this script. I want to use a pre-trained caffe model(e.g. models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel ) to extract features from a set of training images in the libsvm format. How should I specify the path of my images if I use your script?

Thank you in advance for any help!