Python GUI-based driving hazard video labelling.
This labelling tool will allow a user to label "hazardous" frames in videos, similar to how it is done in the UK Hazard Perception Test.
This is an early prototype used primarily for myself in my research and therefore is in a perpetual state of being "in-dev", provided as is.
It's fairly simple and lightweight.
- Clone this repo into a location of your choosing
- Install the dependencies in
requirements.txt, either globally or in a virtual environment
The labelling tool can operate in three ways: files, folder or default.
-
The default mode is run by executing the script with no arguments :
python main.py. This opens the GUI with no pre set video and allows you to choose a video somewhere on your file system. -
In files mode, video filepath(s) are passed at the command line and is loaded sequentially:
python main.py --filepath /path/to/file [/path/to/file2 ...]. After the sequence has terminated, you can choose another video via the GUI. -
Folder mode is similar to the above except that a path to a folder is given, and the tool recursively searches down this folder for all videos. This is the prefered mode. Run by:
python main.py --folder /path/to/folder.
To flag a hazard, simply press the Enter key, or click anywhere on the video.
The labels are saved as .csv files at the location of the video if the Save labels button is pressed. Otherwise, you can click Export at the bottom of the timeline and choose.
The timeline has an upper limit of frames, so flags won't appear visually past this time, though labelling is still happening. This is about 300k frames, so it only occurs when labelling long-ish videos.
Currently, 60 frames after a hazard has been flagged is also flagged. This corresponds to 2 seconds for a 30fps video, and is because hazards are expected to last longer than 1 video frame.
When Next video is pressed, flags are saved at the video location.
The timeline can be undocked form the window by dragging.