VOC Object Detection Scripts

Description

The idea of this repository is to help people generate and test custom datasets for the VOC format in order to be able train Object Detection models.

Usage of gt2voc script

1. Create a file named dataset_gt.gt using the format below: name_of_image x0 y0 x1 y1 class_name
2. then just run the gt2voc.py script following the instructions below:

cd scripts
python gt2voc.py path/to/image/folder/ path/to/image/folder2/ path/to/output/folder/

• the dataset_gt.gt file needs to be inside the output folder.
• The position is in percent of the image (eg: person1 0.2 0.1 0.4 0.5 person)
• the name of the image is without the extension (.jpg,.png)

To run the script it needs to have installed opencv-python and xml.

Additional scripts

The check_xml.py helps to check if the bounding boxes are in the right part of the image

cd scripts
python check_xml.py xml_directory images_directory

The count_objects_xml.py helps to count how many objects each class has for the specific text file with the annotations in order to help generating the test.txt file.

cd scripts
python count_objects_xml.py text_file