Dataset from ftp://motinas.elec.qmul.ac.uk/pub/iLids/

image
Annotations  for 5 segments of the video: AVSS_PV_EVAL.avi from ftp://motinas.elec.qmul.ac.uk/pub/iLids/ Using the LabelMe Video Tool.

The video was segmented in segments of 30 seconds using the script ffsplit.sh AVSS_PV_EVAL.avi 30
We select the video segments: 1, 2, 5, 15, 26.

AVSS_PV_EVAL-001.xml
AVSS_PV_EVAL-002.xml
AVSS_PV_EVAL-005.xml
AVSS_PV_EVAL-015.xml
AVSS_PV_EVAL-026.xml

To read the xmls and transform to PASCAL format, some additional matlab functions were implemented which work with the LabelMe MATLAB ToolBox which are:

readXMLs.m  Read the XML file
deleteRepeted.m Because the system consider all the annotations start at frame 1 to the last frame, for that It repetted the first and last true annotation 1 and to the last frames respectively


For the PASCAL format we consider:   
<ID_Object, ID_Frame, X1, Y1, X2, Y2>

avss_annotation_AVSS_PV_EVAL-001.txt

avss_annotation_AVSS_PV_EVAL-002.txt    

avss_annotation_AVSS_PV_EVAL-005.txt

avss_annotation_AVSS_PV_EVAL-015.txt

avss_annotation_AVSS_PV_EVAL-026.txt

Experimental Results
  • AVSS .... 001.avi
12
34
56
  • AVSS .... 002.avi
78
910
1112
  • AVSS .... 015.avi
1314
1516
1718
  • AVSS .... 026.avi
1020

21222324