![]() |
TI Autonomous Driving Algorithms (TIADALG) Library User Guide
|
Multi task networks are trained to perform multiple tasks on a single inference of the network. Thus it saves on compute and hence suitable for real-time application. Theoretically, for correlated tasks, it further improve performance compared to each individual tasks.
We will discuss an example network that uses two inputs (Optical flow, Current frame) and performs three tasks namely (depth, semantic and motion) for each input pixel. The model takes two inputs and three output as shown in the figure below:
|
© Copyright 2018 Texas Instruments Incorporated. All rights reserved. |
Document generated by doxygen 1.8.6 |