/video_annotations_to_yolo

Conversion utility for Label Studio video annotations to a YOLO-compatible format, including bounding box interpolation

Primary LanguagePython

Label Studio to YOLO Video Annotations Converter

Description

This repository features a Python script that processes video annotations created using Label Studio, converting them into a format suitable for the YOLO (You Only Look Once) object detection model. The script interpolates bounding boxes for each intermediate frame based on key-frame annotations (as needed) and exports the labels (i.e., bounding box coordinates), optionally along with the corresponding frames, into a YOLO-compatible format. As of Label Studio version 1.7.0, such functionality isn't inherently available. Please note that video annotations should be exported in the JSON-MIN format from Label Studio.

Usage

  1. Export your Label Studio video annotations in JSON-MIN format.
  2. Run the provided Python script from the command line as follows: 
    python ls2yolo.py --json_path path_to_your_json_annotations --video_path path_to_your_video_file --output_base path_to_output_directory
    video_path is optional. If not provided, the script only generates YOLO labels without extracting frames from the video.

Please refer to the comments in the Python script for more detailed instructions.