VideoTimelineAnnotator

A tool for video timeline segmentation and annotation.

Script Parameters Description

• -v or --video_path: path to the video file or directory containing video files, required argument.
• -c or --class_cfg: path to the YAML file with event classes, required argument.
• -o or --output_dir: path to the directory for saving annotations; if not provided, annotations are saved in the same directory as the video being annotated.
• -s or --frames_skip: number of frames to skip during playback; if not provided, the default value of 1 is used.

Example Usage

python labeller.py -v data -c classes.yml

In this example, the script will process all video files in the data directory, use event classes from classes.yml, and save annotations in the same directory as the source videos, skipping 10 frames during playback.

python labeller.py -v video.mp4 -c configs/classes.yml -o annotations -s 10

In this example, the script will run with the video.mp4 file, use event classes from configs/classes.yml, and save annotations in the annotations directory, skipping 10 frames during playback.

Usage Instructions

1. Run the labeller.py script with the required parameters.
2. A video window with a timeline will appear. The script can be run on a single video or a directory of videos. In the latter case, the script will process each video sequentially.
3. Use the following keys to navigate the video:
   • Space: start/stop playback
   • a or ← Left Arrow: move one frame backward
   • d or → Right Arrow: move one frame forward
   • w or ↑ Up Arrow: move 10 frames backward
   • s or ↓ Down Arrow: move 10 frames forward
4. Marking the start and end frames can be done in any order (e.g., [start, end] or [end, start]). Events can be annotated in two ways:
   • Using text input for event id:
     1. Pause the video at the start/end frame of the event.
     2. Press Enter.
     3. Enter the event id.
     4. Confirm by pressing Enter again.
     5. Pause the video at the end/start frame of the event (the second marker).
     6. Press Enter again.
   • Using keyboard input for event class id (works with single-character ids):
     1. Pause the video at the start/end frame of the event.
     2. Press the key corresponding to the event id.
     3. Pause the video at the end/start frame of the event (the second marker).
     4. Press Enter.
5. To finish annotation and save the annotations, press the \ key. Annotations will be saved in the file within the directory specified by the -o parameter.
6. To cancel the video annotation, press Esc.

System Requirements

• Python version 3.8 or higher
• Libraries opencv and yaml for Python

You can install the required libraries using:

pip install opencv-python pyyaml