Instance segmentation XXL-CT challenge of an historic airplane

The field of non-destructive testing (NDT) has been expanded in recent years with many new possibilities. Among other things, new methods of image segmentation (through the use of machine learning or artificial intelligence) and big data scenarios are gaining traction. Together with the ADA Lovelace Center of the Fraunhofer IIS, the German Society for Non-Destructive Testing (DGZFP) brings these two developments together as part of an (international) challenge and find an answer to the question:

 

"Which automatic or interactive methods from the areas of digital image processing, machine learning or deep Neural networks can segment parts of an old airplane with the highes quality?"

 

The challenge is divided into several phases (see below). After a training phase in which we provide known pairs of input volumes and there manual annotations, the goal is to segment a dataset for which we will not provide ground truth data. We will then evaluate and compare the results of the submitted segmentations with the manually segmented volumes.

Sementic overview of the data aquisition and segmentation process

Would you like to participate?

Please fill in the form so we can send you the details.  

Bitte füllen Sie das Pflichtfeld aus.
Please fill mandatory field
Please fill mandatory field
Please fill mandatory field

Phases

Phase 0 (Preparation)

  • Website is online
  • Registration is open
  • Example data is provided for download
  • We open a discussion forum for registered participants

Phase 1 (Training)

  • Starts 2022-10-01
  • Each registered participant can download a set of seven 512x512x512 sub-volumes and their corresponding ground-truth segmentation (which may not be redistributed during the duration of the challenge)

Phase 2 (Inference)

  • Starts 2022-11-01 (ends 2022-11-15)
  • Registration closes
  • We provide an additional sub-volume without corresponding ground-truth.
  • Submission period of segmentation and a brief description (1-2 pages) about their utilized methodology.

Phase 3 (Evaluation)

  • Starts 2022-11-15
  • Submission closes
  • We evaluate the submissions
  • We will select and present the best solutions and release the training as well as the ground truth data under a CC-BY-SA licence

Phase 4 (Publication)

  • Start spring 2023
  • A joint publication is created (with all participants as co-authors)

Phase 5 (End)

  • Discussion forum and website closes

Organisation and Administration

ADA Lovelace Center

 

Fraunhofer EZRT

  • Overall: Thomas Wittenberg,
  • Technical: Roland Gruber, Stefan Gerth, Michael Salamon,
  • Marketing: Nadine Chrobok-Pensky, Anikö Enderlein

DGZFP

  • Thomas Wenzel
  • Marika Maniszewski

Deutsches Museum, München

  • Andreas Hempfer

Scientific Committee

  • PD Dr. Thomas Wittenberg (Fraunhofer IIS, FAU Erlangen)
  • Prof. Dr. Thomas Sauer (Univ. Passau, Fraunhofer IIS)
  • Dr. Thomas Wenzel (DGZfP)
  • Prof. Dr. Thorsten Buzug  (Univ. Lübeck, Fraunhofer IMTE)

Literatur

[1] Gruber, R., Gerth, S., Claußen, J., Wörlein, N., Uhlmann, N., Wittenberg, T., 2020. Exploring Flood Filling Networks for Instance Segmentation of XXL-Volumetric and Bulk Material CT Data. Journal of Nondestructive Evaluation 40, https://doi.org/10.1007/s10921-020-00734-w