Advancements in Synthetic Video Generation for Autonomous Driving
The mean score to compute relative distance is calculated using the given formula where μ r and μ g are optical feature vector distances for real and generated frames respectively .
5.3 CONCLUSION
2 2
D KLT = || μ r – μ 2 g ||
• The proposed model could generate realistic scenarios if given different segmented inputs . It would perform better with increasing frames and cascading layers .
• FID and KLT score-based proposed metrics will help us evaluate any video synthesis models and eventually reduce human intervention ( as shown in Table 5-1 ).
• The proposed model helps us generate adversarial data ( as shown in Figure 5.1 ), which eventually helps improve object detection capabilities ( as shown in Table 5-2 ).
REFERENCES
[ 1 ] N . Kalra and S . M . Paddock , Driving to Safety : How Many Miles of Driving Would It Take to Demonstrate Autonomous Vehicle Reliability ? Santa Monica , CA : RAND Corporation , 2016 .
[ 2 ] K . K . Patel , “ A Simulation Environment with Reduced Reality Gap for Testing Autonomous Vehicles ,” University of Windsor , 2020 .
[ 3 ] P . Zhu , R . Abdal , Y . Qin , and P . Wonka , “ SEAN : Image synthesis with semantic regionadaptive normalization ,” Proc . IEEE Comput . Soc . Conf . Comput . Vis . Pattern Recognit ., pp . 5103 – 5112 , 2020 , doi : 10.1109 / CVPR42600.2020.00515 .
[ 4 ] A . Mallya , T . -C . Wang , K . Sapra , and M . -Y . Liu , “ World-Consistent Video-to-Video Synthesis ,” Lect . Notes Comput . Sci . ( including Subser . Lect . Notes Artif . Intell . Lect . Notes Bioinformatics ), vol . 12353 LNCS , pp . 359 – 378 , Jul . 2020 , doi : 10.1007 / 978-3-030-58598- 3 _ 22 .
[ 5 ] T . C . Wang et al ., “ Video-to-video synthesis ,” Adv . Neural Inf . Process . Syst ., vol . 2018- Decem , pp . 1144 – 1156 , 2018 .
[ 6 ] T . C . Wang , M . Y . Liu , A . Tao , G . Liu , J . Kautz , and B . Catanzaro , “ Few-shot video-to-video synthesis ,” arXiv , 2019 .
[ 7 ] M . Liao , F . Lu , D . Zhou , S . Zhang , W . Li , and R . Yang , “ DVI : Depth Guided Video Inpainting for Autonomous Driving ,” arXiv , pp . 1 – 16 , 2020 .
[ 8 ] B . Cheng et al ., “ Panoptic-deeplab : A simple , strong , and fast baseline for bottom-up panoptic segmentation ,” Proc . IEEE Comput . Soc . Conf . Comput . Vis . Pattern Recognit ., pp . 12472 – 12482 , 2020 , doi : 10.1109 / CVPR42600.2020.01249 .
Journal of Innovation 127