Publication detail

Cheap Rendering vs. Costly Annotation: Rendered Omnidirectional Dataset of Vehicles

ŠLOSÁR, P. JURÁNEK, R. HEROUT, A.

Original Title

Cheap Rendering vs. Costly Annotation: Rendered Omnidirectional Dataset of Vehicles

English Title

Cheap Rendering vs. Costly Annotation: Rendered Omnidirectional Dataset of Vehicles

Type

conference paper

Language

en

Original Abstract

Detection of vehicles in traffic surveillance needs good and large training datasets in order to achieve competitive detection rates. We are showing an approach to automatic synthesis of custom datasets, simulating various major influences: viewpoint, camera parameters, sunlight, surrounding environment, etc. Our goal is to create a competitive vehicle detector which "has not seen a real car before." We are using Blender as the modeling and rendering engine. A suitable scene graph accompanied by a set of scripts was created, that allows simple configuration of the synthesized dataset. The generator is also capable of storing rich set of metadata that are used as annotations of the synthesized images. We synthesized several experimental datasets, evaluated their statistical properties, as compared to real-life datasets. Most importantly, we trained a detector on the synthetic data. Its detection performance is comparable to a detector trained on state-of-the-art real-life dataset. Synthesis of a dataset of 10,000 images takes only several hours, which is much more efficient, compared to manual annotation, let aside the possibility of human error in annotation.

English abstract

Detection of vehicles in traffic surveillance needs good and large training datasets in order to achieve competitive detection rates. We are showing an approach to automatic synthesis of custom datasets, simulating various major influences: viewpoint, camera parameters, sunlight, surrounding environment, etc. Our goal is to create a competitive vehicle detector which "has not seen a real car before." We are using Blender as the modeling and rendering engine. A suitable scene graph accompanied by a set of scripts was created, that allows simple configuration of the synthesized dataset. The generator is also capable of storing rich set of metadata that are used as annotations of the synthesized images. We synthesized several experimental datasets, evaluated their statistical properties, as compared to real-life datasets. Most importantly, we trained a detector on the synthetic data. Its detection performance is comparable to a detector trained on state-of-the-art real-life dataset. Synthesis of a dataset of 10,000 images takes only several hours, which is much more efficient, compared to manual annotation, let aside the possibility of human error in annotation.

Keywords

Realistic Rendering, Dataset of Vehicles, Omnidirectional Views, Computer Vision, Object Detection

RIV year

2014

Released

28.05.2014

Publisher

Comenius University in Bratislava

Location

Smolenice

ISBN

978-80-223-3601-7

Book

Proceedings of Spring Conference on Computer Graphics

Edition

NEUVEDEN

Edition number

NEUVEDEN

Pages from

105

Pages to

112

Pages count

8

URL

Documents

BibTex


@inproceedings{BUT111601,
  author="Peter {Šlosár} and Roman {Juránek} and Adam {Herout}",
  title="Cheap Rendering vs. Costly Annotation: Rendered Omnidirectional Dataset of Vehicles",
  annote="Detection of vehicles in traffic surveillance needs good and large training
datasets in order to achieve competitive detection rates. We are showing an
approach to automatic synthesis of custom datasets, simulating various major
influences: viewpoint, camera parameters, sunlight, surrounding environment, etc.
Our goal is to create a competitive vehicle detector which "has not seen a real
car before." We are using Blender as the modeling and rendering engine.
A suitable scene graph accompanied by a set of scripts was created, that allows
simple configuration of the synthesized dataset. The generator is also capable of
storing rich set of metadata that are used as annotations of the synthesized
images. We synthesized several experimental datasets, evaluated their statistical
properties, as compared to real-life datasets. Most importantly, we trained
a detector on the synthetic data. Its detection performance is comparable to
a detector trained on state-of-the-art real-life dataset. Synthesis of a dataset
of 10,000 images takes only several hours, which is much more efficient, compared
to manual annotation, let aside the possibility of human error in annotation.",
  address="Comenius University in Bratislava",
  booktitle="Proceedings of Spring Conference on Computer Graphics",
  chapter="111601",
  doi="10.1145/2643188.2643191",
  edition="NEUVEDEN",
  howpublished="print",
  institution="Comenius University in Bratislava",
  year="2014",
  month="may",
  pages="105--112",
  publisher="Comenius University in Bratislava",
  type="conference paper"
}