IDL SVM Library Malloc Allocation Error_keeps coming

Prev Go to previous topic

Go to next topic

Last Post 03 Jul 2014 01:10 AM by anon

3 Replies

Sort:


You are not authorized to post a reply.

Author

Messages

anon

New Member

Posts:

03 Jul 2014 01:10 AM
Hi,I use ENVI 4.8.I am trying SVM classification on a one-band (a VI- GDVI), with a small coverage (159600 pixels). This error message came; envi_svm_doit: An error has occurred during processingError:"IDL SVM Library Malloc allocation error" The result may be invalid.then I defragmented the working and the root directories, and ran the classification again.but the same error message keeps coming.I tried the same with NDVI, before and after. But it works!!NDVI (range : -0.18 to 0.57), GDVI (range : -0.50 to 0.96).The training data set used was drawn (ROIs) on the multispectral image of the same coverage, 8classes.What could be the problem? How to solve this?Thanks

MariM

Veteran Member

Posts:2396
Veteran Member

03 Jul 2014 08:56 AM
Do any of your ROI classes have a small sample of pixels? I have seen this error when the ROI has only 1 pixel in some cases.

Deleted User

New Member

Posts:

06 Jul 2014 01:32 AM
Thanks. There is a class with 17 polygons/ 42 pixels, which has the smallest no: of pixels. And almost all classes have 1-pixel polygons in the ROI. Anyway, assuming the one-pixel polygons are the problem, I edited the ROI classes not to have a single 1-pixel polygon. Tried SVM again, had no luck.I deleted all the files created using the problematic VI...GDVI file. Then I saved the session to script (re-wrote actually... normally we keep the last session saved to script if we hadn't finished up the task). Then ran the SVM classification on the file again.It was successful.I am guessing some file sharing might have been causing trouble?True...my problem got solved...but what do you think of this? ...good to knowThanks a lot

MariM

Veteran Member

Posts:2396
Veteran Member

10 Jul 2014 09:21 AM
It is hard to tell what might be causing it. It would be good to track down if we could access the dataset. If this is something you can do, please contact Technical Support directly.

You are not authorized to post a reply.