Convert PascalVOC Dataset to YOLO

This scripts converts a PascalVOC dataset to YOLO format. It has the following functionality:

• creates the directory structure needed in order to train a YOLO network
• creates the obj.data, obj.names, train.txt and val.txt files based on the given input
• reads the xml files and converts them to YOLO txt files and saves them to data/obj/
• copy the image files from PascalVOC JPEGImages/ folder to YOLO data/obj/

The PascalVOC dataset was generated with the VOTT annotation and labeling tool from a video recorded and labeled by me and has the following directory structure:

.
├── Annotations
│   ├── test001.mp4#t=0.1.xml
│   ├── test001.mp4#t=0.2.xml
│   ...
├── ImageSets
│   └── Main
│       ├── bus_train.txt
│       ├── bus_val.txt
│       ├── car_train.txt
│       └── car_val.txt
└── JPEGImages
    ├── test001.mp4#t=0.1.jpg
    ├── test001.mp4#t=0.2.jpg
    ...

After running the pascalVOC2YOLO.py script the generated YOLO dataset should be:

.
└── data
    └── obj
    │   ├── test001.mp4#t=0.1.jpg
    │   ├── test001.mp4#t=0.1.txt
    │   ├── test001.mp4#t=0.2.jpg
    │   ├── test001.mp4#t=0.2.txt
    │   ...
    │
    ├── obj.data
    ├── obj.names
    ├── train.txt
    └── val.txt

Where the YOLO txt files under the obj folder have the format:

alx@jnano1:~/dataset-YOLO/data/obj$ cat test001.mp4#t=0.1.txt
1 0.8065568943695005 0.7365158472839091 0.050766507208635285 0.04888481655003832
1 0.08474938686836826 0.7325652625446453 0.04552266480924344 0.04590375467312116
1 0.13165040789512703 0.7152108707094267 0.0375028179792637 0.04116810055598158
1 0.1768199617900546 0.6906309648308799 0.05906558250706307 0.057836839172737686

If you are interested in the YOLO dataset I generated with this script you can download it from here. The dataset is called Roundabout Traffic Dataset since it is based on a recorded video with a fixed camera in a roundabout.

The script was tested on NVIDIA Jetson Nano.

Disclaimer: This code is a modified version of Joseph Redmon's voc_label.py.

How to run

Before running the script you need to edit a few variables.

1. Modify the classes variable at line 8. Make sure to keep the order of classes correct.

classes = ['bus', 'car'] # Add your classes

2. Point where your PascalVOC dataset is by changing VOC_DIR at line 10.

VOC_DIR = '/home/alx/dataset-VOC' # Path to the input PascalVOC dataset 

3. Edit the YOLO_DIR at line 11 to set the output folder.

YOLO_DIR = '/home/alx/dataset-YOLO' # Path to where to save the YOLO dataset

Finally run the script. How long it takes to run depends on your dataset and your environment.

python pascalVOC2YOLO.py