opencv_traincascade "-baseFormatSave" flag for legacy XML file format error

[opencv-pushbot]

Transferred from http://code.opencv.org/issues/2387

|| Joel Mckay on 2012-09-25 17:35
|| Priority: Normal
|| Affected: branch '2.4'
|| Category: objdetect
|| Tracker: Bug
|| Difficulty: None
|| PR: 
|| Platform: None / None

opencv_traincascade "-baseFormatSave" flag for legacy XML file format error

The legacy format is still used for various examples.

opencv_traincascade "-baseFormatSave" flag currently outputs a legacy format XML with the tag:
<pre>
<cascade>
...
</cascade>
</pre>

This will cause an error like:
<pre>
 3 34.0
OpenCV Error: Unspecified error (The node does not represent a user object (unknown type?)) in cvRead, file /home/yourmom/SDK/opencv/trunk/opencv/modules/core/src/persistence.cpp, line 4996
terminate called after throwing an instance of 'cv::Exception'
  what():  /home/yourmom/SDK/opencv/trunk/opencv/modules/core/src/persistence.cpp:4996: error: (-2) The node does not represent a user object (unknown type?) in function cvRead
</pre>

To ensure compatibility the tag must be formatted as the following:
<pre>
<output type_id="opencv-haar-classifier">
...
</output>
</pre>


The opencv version used was: OpenCV 2.4.9 commit:ce4ac2d

Additionally, a newly trained "lost" cascade.xml may be manually edited with this specific tag to enable backward compatibility for GPU examples etc. I am regression testing the cascade, and so far it appears to behave itself amicably. 

Cheers,
J

History

Alexander Smorkalov on 2012-10-03 08:17

-   Target version set to 2.4.3
-   Assignee set to Vadim Pisarevsky

Andrey Kamaev on 2012-10-31 14:09

-   Target version changed from 2.4.3 to 3.0

Marina Kolpakova on 2012-11-02 09:59

-   % Done changed from 80 to 0
-   Assignee changed from Vadim Pisarevsky to Marina Kolpakova
-   Estimated time deleted (1.00)

Marina Kolpakova on 2013-04-05 23:31

-   Affected version set to branch '2.4'
-   Assignee deleted (Marina Kolpakova)

[StevenPuttemans]

This was solved by me and @Dikay900
--> correct models for GPU processing where added in seperate module
--> all the other samples and modules work with the new train cascade format

It can thus be closed.