3D Pose Estimation from Color Images without Manual Annotations

Mahdi Rad, Markus Oberweger, Vincent Lepetit

Research output: Chapter in Book/Report/Conference proceedingChapterResearchpeer-review

Abstract

3D pose estimation is an important problem with many potential applications. However, 3D acquiring annotations for color images is a difficult task. To create training data, the annotating is usually done with the help of markers or a robotic system, which in both cases is very cumbersome, expensive, or sometimes even impossible, especially from color images. Another option is to use synthetic images for training. However, synthetic images do not resemble real images exactly. To bridge this domain gap, Generative Adversarial Networks or transfer learning techniques can be used but, they require some annotated real images to learn the domain transfer. To overcome these problems, we propose a novel approach in this paper. Section II gives a short summary of our approach that uses synthetic data only, and Section III shows some results.
Original languageEnglish
Title of host publicationProceedings of the joint OAGM & ARW Workshop 2019
EditorsAndreas Pichler, Peter M. Roth, Robert Slabatnig, Gernot Stübl
Place of PublicationGraz
PublisherVerlag der Technischen Universität Graz
Number of pages1
ISBN (Electronic)9783851256635
DOIs
Publication statusPublished - 2019
EventARW & OAGM Workshop 2019: Austrian Robotics Workshop and OAGM Workshop 2019 - Steyr, Austria
Duration: 9 May 201910 May 2019

Conference

ConferenceARW & OAGM Workshop 2019
CountryAustria
CitySteyr
Period9/05/1910/05/19

Fingerprint

Color
Robotics

Keywords

  • 3D Object Pose Estimatio
  • 3D Hand Pose Estimation
  • Domain Transfer

Cite this

Rad, M., Oberweger, M., & Lepetit, V. (2019). 3D Pose Estimation from Color Images without Manual Annotations. In A. Pichler, P. M. Roth, R. Slabatnig, & G. Stübl (Eds.), Proceedings of the joint OAGM & ARW Workshop 2019 Graz: Verlag der Technischen Universität Graz. https://doi.org/10.3217/978-3-85125-663-5-27

3D Pose Estimation from Color Images without Manual Annotations. / Rad, Mahdi; Oberweger, Markus; Lepetit, Vincent.

Proceedings of the joint OAGM & ARW Workshop 2019 . ed. / Andreas Pichler; Peter M. Roth; Robert Slabatnig; Gernot Stübl. Graz : Verlag der Technischen Universität Graz, 2019.

Research output: Chapter in Book/Report/Conference proceedingChapterResearchpeer-review

Rad, M, Oberweger, M & Lepetit, V 2019, 3D Pose Estimation from Color Images without Manual Annotations. in A Pichler, PM Roth, R Slabatnig & G Stübl (eds), Proceedings of the joint OAGM & ARW Workshop 2019 . Verlag der Technischen Universität Graz, Graz, ARW & OAGM Workshop 2019, Steyr, Austria, 9/05/19. https://doi.org/10.3217/978-3-85125-663-5-27
Rad M, Oberweger M, Lepetit V. 3D Pose Estimation from Color Images without Manual Annotations. In Pichler A, Roth PM, Slabatnig R, Stübl G, editors, Proceedings of the joint OAGM & ARW Workshop 2019 . Graz: Verlag der Technischen Universität Graz. 2019 https://doi.org/10.3217/978-3-85125-663-5-27
Rad, Mahdi ; Oberweger, Markus ; Lepetit, Vincent. / 3D Pose Estimation from Color Images without Manual Annotations. Proceedings of the joint OAGM & ARW Workshop 2019 . editor / Andreas Pichler ; Peter M. Roth ; Robert Slabatnig ; Gernot Stübl. Graz : Verlag der Technischen Universität Graz, 2019.
@inbook{b82aead13aa2437690a62ce2cddd188e,
title = "3D Pose Estimation from Color Images without Manual Annotations",
abstract = "3D pose estimation is an important problem with many potential applications. However, 3D acquiring annotations for color images is a difficult task. To create training data, the annotating is usually done with the help of markers or a robotic system, which in both cases is very cumbersome, expensive, or sometimes even impossible, especially from color images. Another option is to use synthetic images for training. However, synthetic images do not resemble real images exactly. To bridge this domain gap, Generative Adversarial Networks or transfer learning techniques can be used but, they require some annotated real images to learn the domain transfer. To overcome these problems, we propose a novel approach in this paper. Section II gives a short summary of our approach that uses synthetic data only, and Section III shows some results.",
keywords = "3D Object Pose Estimatio, 3D Hand Pose Estimation, Domain Transfer",
author = "Mahdi Rad and Markus Oberweger and Vincent Lepetit",
year = "2019",
doi = "10.3217/978-3-85125-663-5-27",
language = "English",
editor = "Andreas Pichler and Roth, {Peter M.} and Slabatnig, {Robert } and Gernot St{\"u}bl",
booktitle = "Proceedings of the joint OAGM & ARW Workshop 2019",
publisher = "Verlag der Technischen Universit{\"a}t Graz",

}

TY - CHAP

T1 - 3D Pose Estimation from Color Images without Manual Annotations

AU - Rad, Mahdi

AU - Oberweger, Markus

AU - Lepetit, Vincent

PY - 2019

Y1 - 2019

N2 - 3D pose estimation is an important problem with many potential applications. However, 3D acquiring annotations for color images is a difficult task. To create training data, the annotating is usually done with the help of markers or a robotic system, which in both cases is very cumbersome, expensive, or sometimes even impossible, especially from color images. Another option is to use synthetic images for training. However, synthetic images do not resemble real images exactly. To bridge this domain gap, Generative Adversarial Networks or transfer learning techniques can be used but, they require some annotated real images to learn the domain transfer. To overcome these problems, we propose a novel approach in this paper. Section II gives a short summary of our approach that uses synthetic data only, and Section III shows some results.

AB - 3D pose estimation is an important problem with many potential applications. However, 3D acquiring annotations for color images is a difficult task. To create training data, the annotating is usually done with the help of markers or a robotic system, which in both cases is very cumbersome, expensive, or sometimes even impossible, especially from color images. Another option is to use synthetic images for training. However, synthetic images do not resemble real images exactly. To bridge this domain gap, Generative Adversarial Networks or transfer learning techniques can be used but, they require some annotated real images to learn the domain transfer. To overcome these problems, we propose a novel approach in this paper. Section II gives a short summary of our approach that uses synthetic data only, and Section III shows some results.

KW - 3D Object Pose Estimatio

KW - 3D Hand Pose Estimation

KW - Domain Transfer

U2 - 10.3217/978-3-85125-663-5-27

DO - 10.3217/978-3-85125-663-5-27

M3 - Chapter

BT - Proceedings of the joint OAGM & ARW Workshop 2019

A2 - Pichler, Andreas

A2 - Roth, Peter M.

A2 - Slabatnig, Robert

A2 - Stübl, Gernot

PB - Verlag der Technischen Universität Graz

CY - Graz

ER -