Loading [MathJax]/extensions/MathMenu.js
Toward a Methodology for Training with Synthetic Data on the Example of Pedestrian Detection in a Frame-by-Frame Semantic Segmentation Task | IEEE Conference Publication | IEEE Xplore

Toward a Methodology for Training with Synthetic Data on the Example of Pedestrian Detection in a Frame-by-Frame Semantic Segmentation Task


Abstract:

In order to make highly/fully automated driving safe, synthetic training and validation data will be required, because critical road situations are too divers and too rar...Show More

Abstract:

In order to make highly/fully automated driving safe, synthetic training and validation data will be required, because critical road situations are too divers and too rare. A few studies on using synthetic data have been published, reporting a general increase in accuracy. In this paper, we propose a novel method to gain more in-depth insights in the quality, performance, and influence of synthetic data during training phase in a bounded setting. We demonstrate this method for the example of pedestrian detection in a frame-by-frame semantic segmentation class.
Date of Conference: 28-28 May 2018
Date Added to IEEE Xplore: 02 September 2018
ISBN Information:
Conference Location: Gothenburg, Sweden

1 Introduction

Taking into consideration the manifold activities in this area, it can be regarded as an accepted fact in the community that synthetic data is required in order to make AI for autonomous driving safe. Critical road situations are too rare and too divers - we cannot find enough real data for training or validating, even when collecting another few million miles. Road situations apparently follow a longtail distribution: relatively few situations happen very often (like “car approaching from the right at an intersection”), while a large number situations happen only rarely (like “child running in front of car”). Also, for many critical situations (like “child running in front of car”) we do not even want to collect real data.

References

References is not available for this document.