ClassifierVOC_for_ClassifulNet

This MATLAB program can devide sample images into servel classes according groundtruth message in XML file

---

PREPARED

VOC2007 dataset is need:

VOC2007

Annotations

name.xml

JPEGImages

name.jpg

labels

name.txt

name.txt can be genarated by code in darnet/scripts/voc_label.py contains:

class_ind(start with 0) xmin xmax ymin ymax

---

START

1. run the lateast main*.m, for example

run main9.m

• Then the result images will be ploted in output1 file which will be named as ：

classes - subclasses - amounts of images.jpg 

• And the result txt will be saved in output4 file which will be named as :

classes - subclasses.txt

• This txt can be edited according to the result images and the code can read these to cover the old result.

---

2. run the printresult.m

• Then the bounding box can be genarated according to the lateast classes result, which will be drawn in output2 file and named as :

new classes - 12.jpg

• annotation_bbx.txt and classifier.txt will also be genarated.

  • annotatioin_bbx.txt :

    class_index(start with 1) tag_index(start with 1) xmin ymin xmax ymax
    

  • annotatioin_bbx.txt :

    imagename class_index change_index
    
    

    change_index above is the method to deal with the images:

    change_index | deal_method
    ---|---
    1 | Original
    2 | Flip Horizontal
    3 | Flip vertical
    4 | Flip Diagonal
    

---

3. The bounding box can be optimizer again with lateast again_optimizer*.m

run again_optimizer2.m

• The new class result can be genarated in output3 file and named as ：

again-new class.jpg