Auto Video Labeling

Data is a huge factor in deep learning algorithms. The larger our data size, the better our model can generalize and learn. However, data preparation is a very laborious and time-consuming process. That's why I wanted to develop an application that I thought would make this stage easier. By using image processing techniques, it can track an object of your choice to a certain extent and saves the image and .txt file to the folder during tracking. Currently, it only works for one class and you can only label one object.

Note: it does not work very stable in videos with multivariate background

You can download the application using this link

Download - auto_video_labeling.exe

1. Open the application

You can also run it like this python main.py

For ubuntu

requirements --> opencv-contrib-python==4.2.0.34

You should run it like this python3 auto_label.py --video video.mp4 --className label --perFrame 1 --classId 0

2. Fill in the relevant fields. If you leave it blank, the default values will be accepted.

Load video path : Specify the video path you want to label. If you leave it blank, your camera will open.

Save per frame: Saved to one file per frame based on the value entered

Class Name : Specifies the folder name and label name to be saved. Note: It cannot contain Turkish characters.

Class Id : label id must be entered

Hotkeys

+--------------------+----------------------+
| s           | Create a rect box           |
+--------------------+----------------------+
| c           | Cancel selection rect box   |
+--------------------+----------------------+
| z           | cancel tracking operation   |
+--------------------+----------------------+
| q           | close the video             |
+--------------------+----------------------+

You can see how the program works in the gif below.

Outputs

• A folder is created according to the class name you entered.

• Labels and images are saved according to the beginning of the name you enter

• Labels are saved in Yolo format.