kzwkt / wnd-charm

Automatically exported from code.google.com/p/wnd-charm
0 stars 0 forks source link

Crash when using color statistics ("-c") #28

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago
Issue filed by Jimmy Tang at Stanford:

     ... for some reason I'm getting seg faults (I'm running wnd-charm on Ubuntu) when I try and do the training. I've attached a "testset" folder with a folder of images and a file for training. The syntax I use is:
     wndchrm train -mlc testset/trainfile.txt testset/globsa.fit
     Do you mind trying this on your system and see if you get the same error? Ideally, this would give me a globsa.fit file on which I can run "wndchrm test globsa.fit report.html" to generate a report file.

colettace response:

I was able to reproduce the the seg fault using a release build, but when I 
hooked up the debugger (gdb) I could not reproduce the seg fault under any 
circumstances. So we know this one is a doozy. Then I tested which command line 
parameters were the source of the crash and it looks like it's the color 
features. We don't really use color statistics at all in our lab, so it's not 
surprising that this is the culprit. So as a work around, I'd suggest avoiding 
using color statistics, or if this is not an option, you can use wndcharm while 
hooked up to a debugger like gdb. It's very simple, just from your bash shell 
type "gdb wndchrm", then type "run train -mlc testfiles.txt outfile.fit" and 
it'll take longer but it won't seg fault.

I'll let you know when we've fixed the bug. I'm filing a bug on our google code 
page and either I or Ilya will respond to it. 

Cheers,
Chris

Test results: (run on octocore opteron running CentOS 5.5)

release build 184:
`wndchrm train -mlc` - seg fault 
`wndchrm train -mc ` -seg fault
`wndchrm train -ml` -completed successfully
`wndchrm train -mlc -d95` - seg fault
`wndchrm train -mc -d95` -seg fault
`wndchrm train -ml -d95` -completed successfully

Original issue reported on code.google.com by christop...@nih.gov on 14 Apr 2011 at 7:29

GoogleCodeExporter commented 9 years ago

Original comment by christop...@nih.gov on 14 Apr 2011 at 7:30

Attachments:

GoogleCodeExporter commented 9 years ago
gdb on a coredump reported segfault in color_value():
647    while (color_functions[FunctionCounter].color>=0)

Was overrunning the color_functions array because it wasn't terminated properly 
in LoadColorsFunctions().
replaced return() with break to catch the array termination code at the end of 
LoadColorsFunctions().

Original comment by i...@cathilya.org on 15 Apr 2011 at 6:57

GoogleCodeExporter commented 9 years ago

Original comment by i...@cathilya.org on 15 Apr 2011 at 6:59