PascalVOC2YOLO

This is the simplest method of converting Pascal VOC annotation file into Darknet YOLO annotation file

To convert the annotation file, place all the Pascal VOC annotation files in one folder. In terminal/command prompt, run the following line of codes

python PascalVOC2YOLO.py -a path_of_the_Pascal_VOC_annotation_folder

OR

if you want the non-normalize location, height, and width of the boundin box, run the following command

python PascalVOC2YOLO.py -a path_of_the_Pascal_VOC_annotation_folder -m nn

The output YOLO annotation files will be saved on the path_of_the_Pascal_VOC_annotation_folder.

Requirements:

Python 3.X
argparse

To install argparse

pip install argparse

Pascal VOC formate

Pascal VOC stores annotation in XML file. Below is an example of Pascal VOC annotation file for object detection.

Pascal VOC Example

<annotation> 
  <folder>Train</folder> 
  <filename>01.png</filename>      
  <path>/path/Train/01.png</path> 
  <source>  
    <database>Unknown</database> 
  </source>
  <size>  
    <width>224</width>  
    <height>224</height>  
    <depth>3</depth>   
  </size> 
  <segmented>0</segmented> 
  <object>  
    <name>36</name>  
    <pose>Frontal</pose>  
    <truncated>0</truncated>  
    <difficult>0</difficult>  
    <occluded>0</occluded>  
    <bndbox>   
      <xmin>90</xmin>   
      <xmax>190</xmax>   
      <ymin>54</ymin>   
      <ymax>70</ymax>  
    </bndbox> 
  </object>
</annotation>

YOLO Darknet:

The label file specifications are:

One row per object Each row is

class x_center y_center width height

Box coordinates must be in normalized xywh format (from 0 - 1). Class numbers are zero-indexed (start from 0).

YOLO Darknet Example

0 0.409 0.225 0.134 0.113
0 0.406 0.266 0.068 0.135
0 0.406 0.256 0.068 0.156
1 0.437 0.279 0.023 0.119
1 0.47 0.279 0.089 0.119