Instance Segmentation XXL-CT Challenge of a Historic Airplane

TL;DR

The ‘Instance Segmentation XXL-CT Challenge of a Historic Airplane’ was conducted to explore automatic or interactive instance segmentation methods for an efficient delineation of the different aircraft components.

Abstract

Instance segmentation of compound objects in XXL-CT imagery poses a unique challenge in non-destructive testing. This complexity arises from the lack of known reference segmentation labels, limited applicable segmentation tools, as well as partially degraded image quality. To asses recent advancements in the field of machine learning-based image segmentation, the "Instance Segmentation XXL-CT Challenge of a Historic Airplane" was conducted. The challenge aimed to explore automatic or interactive instance segmentation methods for an efficient delineation of the different aircraft components, such as screws, rivets, metal sheets or pressure tubes. We report the organization and outcome of this challenge and describe the capabilities and limitations of the submitted segmentation methods.

Figures (13)

• Figure 1: Clipped 3D-rendering of the XXL-CT reconstruction of the front part of the historical Me 163 airplane containing the seven training sub-volumes $V_1, \dotsc, V_7$ () as well as the sub-volume $V_{\text{test}}$ () used for the test phase of the challenge.
• Figure 2: Renderings of sub-volume $V_\text{test}$ (3072, 6656, 1024). While Figure \ref{['fig:description-example-volume']} shows the unannotated sub-volume, Figure \ref{['fig:description-example-all']} depicts all manually labelled segments separated by colour. To increase clarity, only the segments of a specific category are shown in the following sub-figures: Figure \ref{['fig:description-example-metallSheet']} shows all metal sheets; Figure \ref{['fig:description-example-pipes']} depicts the presumably pressure-carrying pipes, pressure tanks and lines; Figure \ref{['fig:description-example-rivetsAndBolts']} contains all rivets and screw connections; Figure \ref{['fig:description-example-mountingAndMiscellaneous']} shows all brackets, clamp connectors, and other miscellaneous transition elements that could not otherwise be assigned a category.
• Figure 3: Flow graph of participant registration and participation over the course of the challenge. From left to right time. Each colour coded line represents one participant: submitted green () ending in checkmark; withdrew red () ending in cross; no response blue () with horizontal line ending.
• Figure 4: The proposed 3D instance segmentation pipeline by Team One which combines the conditional detection transformer (CDETR) for 2D instance segmentation with subsequent 3D-matching.
• Figure 5: Correlation matrix of the segmentation of sub-volume $V_\text{test}$ (3072, 6656, 1024) between the reference annotation and the result from Team One Figure \ref{['fig:correlationMatrix-engster']} and Team Two Figure \ref{['fig:correlationMatrix-michen']}. Figures \ref{['fig:correlationMatrix-engster-connectedComponent-mainDiagonal']} and \ref{['fig:correlationMatrix-michen-connectedComponent-mainDiagonal']} show the correlation matrices after postprocessing and limited to the main diagonal for Team One and Team Two respectively.