Bringing Creativity, Agility, and Efficiency with Generative AI in Industries 24th Edition | Page 111

Advancements in Synthetic Video Generation for Autonomous Driving
CONTENTS 1 Overview ...................................................................................................................... 108
2 Motivation .................................................................................................................... 109 2.1 Survey ............................................................................................................................. 109
3 Proposals and Contributions .......................................................................................... 117
4 Methodology ................................................................................................................ 118
4.1
Generative Adversarial Networks ..................................................................................... 118
4.2
Training of Generative Adversarial Networks ................................................................... 118
4.3
Objective Functions in Commonly used GANs ................................................................... 118
4.4
Generator Architecture .................................................................................................... 119
4.5
Discriminator Architecture ............................................................................................... 121
4.6
Objective Functions Used in Proposed GAN ...................................................................... 122
4.7
Training and Validation .................................................................................................... 123
5 Results and Conclusions ................................................................................................ 124
5.1
Sample Frame Output ...................................................................................................... 124
5.2
KPI ( Key Performance Indicators ) Evaluation .................................................................... 125
5.3
Conclusion ....................................................................................................................... 127
6 References .................................................................................................................... 127 7 Acknowledgements ....................................................................................................... 129
FIGURES
Figure 2-1 : Different environmental conditions clockwise from top-left daylight , rain scene , sunlight , evening . [ 15 ] ..................................................................................................................................... 110
Figure 2-2 : Depth and segmentation output from CARLA simulator . [ 15 ]................................................ 110 Figure 2-3 : Architecture for low resolution outputs . [ 5 ] ........................................................................... 111
Figure 2-4 : Architecture for high resolution outputs G1 corresponds to lower architecture network and G2 corresponds to higher resolution network . [ 5 ] ........................................................................... 111
Figure 2-5 : Architecture of label / flow embedding , image and segmentation used in world-consistent video-to-video synthesis . [ 4 ] ............................................................................................................. 112
Figure 2-6 : Architecture of generator used in world-consistent video-to-video synthesis . [ 4 ] ................ 113
Figure 2-7 : Architecture of multi SPADE residual block used in world-consistent video-to-video synthesis . [ 4 ] ..................................................................................................................................... 113
Figure 2-8 : Instance as well as label used for encoding . [ 12 ] .................................................................... 114 Figure 2-9 : SPADE normalization . [ 13 ] ....................................................................................................... 115
106 March 2024