Limitations for Trainingset?

asked 2016-04-28 09:59:09 -0500

Schinkenbrot gravatar image

updated 2016-04-28 10:25:23 -0500

berak gravatar image

Hi! Are there limitations for de size of the training set? When I want to train a bigger training set, I obtain this error :

System.Runtime.InteropServices.SEHException msvcr120d.dll!_CxxThrowException(void * pExceptionObject, const _s__ThrowInfo * pThrowInfo) Zeile 154 C++

I am also not able to load an previous created Eigenface- Recognizer with another Trainingset ( //model->load("face-rec-model-eigen.xml"); ) . Same error. That´s strange because it works with my Fisherface-Recognizer.

I have no idea what´s wrong. Thanks for your help.

edit retag flag offensive close merge delete

Comments

how many images ? do you simply run out of memory ? line 154 of what ? does the xml you write, exceed a certain size ?

berak gravatar imageberak ( 2016-04-28 10:15:10 -0500 )edit

I think I simply run out of memory. I decreased the solution of the images (282) and the training worked. After that, I wanted to train my old training set with a lower solution (114 images) but that failed (same error as above). Now nothing works! Even to start my application takes forever. Is it possible to clear the memory? Or what could I do? (Sorry if that is a stupid question, but I am not a programmer and never had such problems) I work with Visual Studio 2013

Schinkenbrot gravatar imageSchinkenbrot ( 2016-04-28 11:59:14 -0500 )edit
  • how large are those images ? imho, you're save to resize them to 100x100, like that you should be able to train on ~10k images
  • for large models, reading it back from xml might take almost as long as retraining ;(
  • "Is it possible to clear the memory?" -- what do you mean here ?
  • are you able to debug it ? (it will take even longer, but at least you'll get some hint, at which point it fails)
berak gravatar imageberak ( 2016-04-28 12:07:25 -0500 )edit

• how large are those images ? My training set uses 146 images with 250x250

• "Is it possible to clear the memory?" I thought there is something like a memory cache which I could clear and solve the problem. I have no idea haha :D

• are you able to debug it ? (it will take even longer, but at least you'll get some hint, at which point it fails) I fails at model->train .... in inline Mat asRowMatrix ... at src.getMat(i).clone().reshape(1,1).convertTo(xi, rtype, alpha, beta);

I have no clue :(

Schinkenbrot gravatar imageSchinkenbrot ( 2016-04-28 13:24:36 -0500 )edit