Showing 1–2 of 2 results for author: Carmignato, S
-
Fully Convolutional Deep Network Architectures for Automatic Short Glass Fiber Semantic Segmentation from CT scans
Authors:
Tomasz Konopczyński,
Danish Rathore,
Jitendra Rathore,
Thorben Kröger,
Lei Zheng,
Christoph S. Garbe,
Simone Carmignato,
Jürgen Hesser
Abstract:
We present the first attempt to perform short glass fiber semantic segmentation from X-ray computed tomography volumetric datasets at medium (3.9 μm isotropic) and low (8.3 μm isotropic) resolution using deep learning architectures. We performed experiments on both synthetic and real CT scans and evaluated deep fully convolutional architectures with both 2D and 3D kernels. Our artificial neural ne…
▽ More
We present the first attempt to perform short glass fiber semantic segmentation from X-ray computed tomography volumetric datasets at medium (3.9 μm isotropic) and low (8.3 μm isotropic) resolution using deep learning architectures. We performed experiments on both synthetic and real CT scans and evaluated deep fully convolutional architectures with both 2D and 3D kernels. Our artificial neural networks outperform existing methods at both medium and low resolution scans.
△ Less
Submitted 4 January, 2019;
originally announced January 2019.
-
Reference Setup for Quantitative Comparison of Segmentation Techniques for Short Glass Fiber CT Data
Authors:
Tomasz Konopczyński,
Jitendra Rathore,
Thorben Kröger,
Lei Zheng,
Christoph S. Garbe,
Simone Carmignato,
Jürgen Hesser
Abstract:
Comparing different algorithms for segmenting glass fibers in industrial computed tomography (CT) scans is difficult due to the absence of a standard reference dataset. In this work, we introduce a set of annotated scans of short-fiber reinforced polymers (SFRP) as well as synthetically created CT volume data together with the evaluation metrics. We suggest both the metrics and this data set as a…
▽ More
Comparing different algorithms for segmenting glass fibers in industrial computed tomography (CT) scans is difficult due to the absence of a standard reference dataset. In this work, we introduce a set of annotated scans of short-fiber reinforced polymers (SFRP) as well as synthetically created CT volume data together with the evaluation metrics. We suggest both the metrics and this data set as a reference for studying the performance of different algorithms. The real scans were acquired by a Nikon MCT225 X-ray CT system. The simulated scans were created by the use of an in-house computational model and third-party commercial software. For both types of data, corresponding ground truth annotations have been prepared, including hand annotations for the real scans and STL models for the synthetic scans. Additionally, a Hessian-based Frangi vesselness filter for fiber segmentation has been implemented and open-sourced to serve as a reference for comparisons.
△ Less
Submitted 4 January, 2019;
originally announced January 2019.
