Marker

A small python tool I wrote to label videos for experiment videos. Given an video file, you can labelled each frame as 0 - Negative, 1 - Questionable, and 2 - Positive. With some small tweaks you can use it to label as many state as you wish.

Usage - Marker

Ensure installation of the required packages by running pip install -r requirements.txt.

pip install -r requirements.txt

To start marking, run:

python mark.py your_video.mp4

Once the video is fully lablled, expect 3 files to be generated:

• *_labels.csv: containing the lablled results.
• *_review.mp4: a review video with the overview progress bar and a cursor.
• *_progress_bar.csv: A pregress bar used to generate review videos.

Manual

• [0,1,2]: Press to label current frame as one of the 3 states, hold to play the video while labelling all viewed frames as such state.
• space: play the video without labelling.
• left/right arrow key: go back/forward 1 frame
• [<,>]: fast forward / rewind 5 seconds of video
• q: quit.

There's two progress bar in the bottom of the interface:

• top being an overview: from the start frame to the end frame, grey - not labelled yet, red/yellow/green = neg/question/positive.
• bottom being a finer resolution one, showing states of the +/- 5 seconds frames that you're currently viewing.

Once you lablled all frames / hit quit, a review video would generate with the overview progress bar and a cursor so you can visually review the corresponding states.

Labelled results are stored in a CSV with 3 columns: frame #, state #, and state alias. In the uploaded version for my experiment, alias are "not flying", "transition", and "flying".

Usage - combine

This is a helper script I wrote. Marker outputs csv per video, but most of the time you'd like multiple videos to be combined in one larger column. I wrote this script to recursively scans within a directory, and combine all outputs into one.

Two csvs would be outputted combined.csv where each video/csv is a column and summary where some primary statistics are ran and each video is served as a row. It can be easily changes by transposing the output csv/dataframe.

---

This script works as intended. If I were to touch this code again I'd probably just make it more configurable or have a GUI interface.