5.7. Semantic Segmentation

By utilizing the Mask2Former model and the PP-LiteSeg model, semantic segmentation is performed on the image to generate segmentation masks for the targets.
The size of the model can have varying effects on the results, particularly in terms of detail.
Below demonstrates the effects exhibited by models of different sizes.

5.7.1. mask2former

The computational platforms currently supported by this network are listed in the table below:

Computing Platform	General X86 Architecture Computers	Radxa ROCK5B+	OrangePi 5 Ultra	NVIDIA Jetson Orin Nano Super	NVIDIA Jetson AGX Orin	D-Robotics RDK X5
Computing Platform	●	●	●	●	●	●
Model Size	216M
Sensor	AC1

The following video demonstrates the offline semantic segmentation effect of this network on a supported computing platform for an open road scenario.

Video Data: open road
Source Code: AC1 Semantic Segmentation mask2former

5.7.2. PPliteseg

The computational platforms currently supported by this network are listed in the table below:

Computing Platform	General X86 Architecture Computers	Radxa ROCK5B+	OrangePi 5 Ultra	NVIDIA Jetson Orin Nano Super	NVIDIA Jetson AGX Orin	D-Robotics RDK X5
Computing Platform	●	●	●	●	●	●
Model Size	49M
Sensor	AC1

The following video demonstrates the offline semantic segmentation effect of this network on a supported computing platform for an open road scenario.

Video Data: open road
Source Code: AC1 Semantic Segmentation PPliteseg