"This site requires JavaScript to work correctly"

Martin Leipert, M.Sc.

Academic Staff

ITC2 0.12

0991/3615-118


Dear visitors, you can find all my publications and my academic CV on my profile page at Pattern Recognition Lab: https://lme.tf.fau.de/person/leipert/


Sortierung:
Journal article
  • Martin Leipert
  • Gabriel Herl
  • J. Stebani
  • Simon Zabler
  • A. Maier

Three Step Volumetric Segmentation for Automated Shoe Fitting.

In: e-Journal of Nondestructive Testing (12th Conference on Industrial Computed Tomography (iCT) 2023, 27 Feb-2 Mar 2023; Fürth, Germany) vol. 28 pg. 1-10

  • (2023)

DOI: 10.58286/27736

This work presents a three-step segmentation process based on Convolutional Neural Networks. The task is to identify the different parts of shoes from Computed Tomography scans of boxed pairs of shoes. The first step of the three-step algorithm uses a scaled-down volume image to separate the shoe material from its surroundings. The second step segments the shoe's inside volume, i.e. the space enclosed by shoe material. The third and last step splits the segmented shoe material into individual components: shoe upper material, outer and insole. The complete process employs CNNs derived from three-dimensional UNets. Residual SE UNet, Dense UNet, and Bottleneck Residual UNet are evaluated for the three steps. The architectures are modified for large receptive fields. The networks are trained and tested for each step separately and conjointly on CT scans comprising various shoe types. The test results inspire hope for using the process for automated segmentation and extraction of meshes from large batches of CT scans. In particular, the first step using a Residual SE UNet achieves an F1-score of 88.2 % for shoes and 58.9 % for the packing material. The second step segments the inside volume with an F1-score of 81.0 %. The third step segments the shoe into its components and achieves an F1-score for insole of 79.5 %, outer sole of 88.7 % and upper material of 81.3 %.
Contribution
  • Martin Leipert
  • Gabriel Herl
  • M. Müller
  • J. Messkemper
  • Simon Zabler

Automated Shoe Metrology by X-Ray Computed Tomography.

In: Proceedings of 3DBODY.TECH 2023 - 14th International Conference and Exhibition on 3D Body Scanning and Processing Technologies.

  • Eds.:
  • N. DApuzzo

Hometrica Consulting - Dr. Nicola D'Apuzzo Ascona, Switzerland

  • (2023)

DOI: 10.15221/23.48

Journal article
  • Martin Leipert
  • Gabriel Herl
  • Simon Zabler
  • A. Maier

Volumetric Instance Detection for Overlapping Shoes in Computed Tomography.

In: e-Journal of Nondestructive Testing (13th Conference on Industrial Computed Tomography (iCT) 2024, 6-9 Feb 2024; School of Engineering, Wals Campus, Austria) vol. 29

  • (2024)

DOI: 10.58286/29229

This work presents an implementation of the Fully Convolutional One-Stage Object Detection (FCOS) object detector for anchorfree detection in volumetric Computed Tomography (CT) data. We test the implementation on a one-class dataset of shoes in closed packages, which is especially challenging to segment due to bounding boxes that overlap up to around 60 %. This complex dataset is tackled with a problem-specific loss function and loss computation. With a Residual-Squeeze-and-Excitation-Network, a lightweight backend is used, and with this, we reach a large receptive field of 2563 voxels. Our algorithm finds instances in volumetric data: for the detection, we reach an average Intersection over Union (IoU) of the predicted bounding box with the ground truth of 0.580 and an Average Precision (AP@IoU=0.5 ) of 84.4% at the best setting. With these results, we can proceed towards instance detection and voxel-wise instance segmentation in volumetric CT data.