[Lush-users] GBLearn2 datasources problem(dsource-image)

[Serg B. <ser...@gm...>]

Hi, All.

I'm trying to use GBLearn2 library for OCR.

So I've got the training working correctly with MNIST-formatted
database(using the dsource-mnist class provided in the demo).
But got some problems with the testing stuff:
I'm trying to use dsource-image, built inside dsource-idx3l-narrow class for
loading the test samples(as in the LeNet example):

      (defvar datasource (new dsource-image 0 0.01) )
      (defvar files_list (list))
      (defvar labels_list (list))
      (for (i 6 (length argv))
           (progn
             (setq files_list (list-merge files_list (list (concat-fname
(nth argv i)))))
             (setq labels_list (list-merge labels_list (list i)))
             ))
      (==> datasource load-pgms files_list labels_list)

      (defvar datasource_idx3l
        (new dsource-idx3l-narrow datasource (length files_list) 0))

But when I do
(==> thetrainer test-sample datasource_idx3l testmeter 1)
i'm getting:
*** lisp_c runtime error: Index dimension does not match size and step
** in: C_fprop_C_c_layer
** in: C_fprop_C_net_cscscf
** in: C_fprop_C_idx3_supervised_module

*** ==> : Run-time error in compiled code


It looks like dsource-image is returning the data in some incorrect format.

Any ideas?