vid2train is a tool that enables users to create image datasets from video files for AI training purposes. It simplifies and streamlines the process of extracting and transforming frames from videos, which can otherwise be a tedious and error-prone task.

vid2train provides the following features:

• It supports any video file format as input and outputs a folder of images with unique names and timestamps.
• It allows users to customize the image size in pixels, with a default value of 512x512.
• It filters out images that are too dark or too bright, to improve the quality and diversity of the dataset.
• It lets users adjust the extraction rate based on a frames per second (FPS) parameter, to influence the number and variety of images in the dataset.

To use vid2train, you need to have Python 3 installed on your system. You also need the following libraries: cv2, numpy, PIL, and torchvision.

python3 vid2train.py

optional arguments:
  -h, --help                  show this help message and exit
  --input INPUT               Input video path.
  --output OUTPUT             Output directory path.
  --crop_size CROP_SIZE       Image crop size in px.                      
  --bright_min BRIGHT_MIN     Minimum brightness level.
  --bright_max BRIGHT_MAX     Maximum brightness level.                  
  --fps FPS                   Images per second.

You can also skip all the flags and enter your input video path when prompted. A new output folder with the same name as your video file + timestamp will be created and the default parameters will be applied:

- output: same directory as file named {filename}{timestamp}
- crop_size: 512
- bright_min: 20
- bright_max: 236
- fps: 1

The program will then ask you to confirm your settings before proceeding with the extraction and transformation process. After, you will see a message indicating how many images were saved in the output directory. You can then use these images for your AI training purposes.

If you have any issues, suggestions, or questions about this script, please feel free to open an issue.

This script is provided "as is" without any warranty. The author is not responsible for any damage or loss caused by using this script. See License for more info.

To acknowledge the use of vid2train in your research or project, please cite the following reference in your publications:

Rich K. MLS (2023). vid2train: A Software Tool for Extracting Images from a Video for Artificial Intelligence Training. GitHub repository. Available at: https://github.com/richkmls/vid2train