[3G1-OS-14a] OS-14

9:00 AM - 9:15 AM JST(12:00 AM - 12:15 AM UTC)

[3G1-OS-14a-01] Supervised CycleGAN for Generating Third-Person View from First-Person View

〇Hiroki Tomura¹, HIRONORI HIRAISHI¹ (1. Ashikaga University)

Keywords:

Image Generation,CycleGAN,Viewpoint Conversion,Driving footage,Driving evaluation

This research aims to evaluate a driver's everyday driving conditions using driving footage from devices such as dashcams. Dashcams provide first-person perspective footage of the area in front of the driver's seat. However, to evaluate one's own driving, a driver can more accurately grasp the vehicle's movements by viewing third-person perspective footage taken from behind. This requires the cooperation of a vehicle driving behind and is not possible with a vehicle alone. To solve this problem, we use a supervised CycleGAN, which incorporates ContextualLoss into CycleGAN, a generative adversarial network model, to train and generate third-person perspective footage from first-person footage taken with the dashcam. We also evaluate the learning model using the dashcam footage.