-
Notifications
You must be signed in to change notification settings - Fork 23
Home
Welcome to the DMD (Driver Monitoring Dataset) repository wiki!
Here you can find information related to the DMD dataset and the available tools to handle the video data and annotations.
Information from the dataset and downloading options are available in the official DMD website.
The dataset is distributed following a defined file structure. We use this structure in all the different tools used to handle de dataset. For a complete description fo the DMD file struct visit this page.
On this page you can learn how to export depth data from the DMD. Also, apply a color map to the images so you can better visualize them.
A tool to annotate temporal actions (TaTo) was developed to annotate temporal actions in the DMD dataset.
The DMD annotations are in ASAM OpenLabel description format; format based on a JSON Schema. To create annotations, the VCD (Video content description) library was used. It can include descriptions of actions, objects, relationships, events, all in one file. This library is compatible with the standard ASAM OpenLabel labelling format. More information about VCD can be found here.
The DMD will have one file per dimension (distraction, drowsiness, gaze and hands). In the case of doing the annotation in another format, they must be temporal annotations, meaning they are time intervals or can be derived into time intervals. This way, a converter to OpenLabel can be built after the annotation process finishes.
To perform the temporal action annotation of the DMD we have defined an annotation criteria which has to be followed by all annotators in order to guarantee the production of consistent annotations.
Check the documents for more details:
-
Distraction: DMD Distraction-related actions criteria
-
Drowiness: DMD Drowsiness-related actions criteria