# A reusable pipeline for large-scale fiber segmentation on unidirectional fiber beds using fully convolutional neural networks: supplementary data
## Alexandre Fioravante de Siqueira1,2, Daniela Mayumi Ushizima1,2, Stéfan J. van der Walt1,2
1 _Berkeley Institute for Data Science, University of California, Berkeley, USA_
2 _Lawrence Berkeley National Laboratory, Berkeley, USA_
* This paper is available on: [[arXiv]](http://arxiv.org/abs/2101.04823)
* Supplementary material available on: [[GitHub]](https://github.com/alexdesiqueira/fcn_microct/)
* Supplementary data available on: [[Dryad]](https://doi.org/10.6078/D1069R)
In this study we separate fibers in ex-situ X-ray CT fiber beds of nine samples from [Larson et al (2019)](http://dx.doi.org/doi:10.18126/M2QM0Z)'s data. To separate the fibers in these samples, we tested four different fully convolutional neural networks — U-net, 3D U-net, Tiramisu, and 3D Tiramisu. When comparing our neural network approach to Larson et al. results, we obtained [Dice](https://doi.org/10.2307/1932409) and [Matthews](https://doi.org/10.1016/0005-2795(75)90109-9) coefficients greater than 92.28 ± 9.65%, reaching up to 98.42 ± 0.03%. This shows that the network results are close to the human-supervised ones in these fiber beds, in some cases separating fibers that the original study could not find. Here you find the data resulting from this study.
## Files containing results of prediction in Larson et al.'s samples
We used twelve different datasets from [Larson et al (2019)](http://dx.doi.org/doi:10.18126/M2QM0Z) in our study. We kept the same folder identifiers used in their original data, for fast cross-reference. The following table sets a correspondence between the name of the folder and the name we refer to that sample in the manuscript.
| Name of the folder | Sample in the manuscript |
| ------------------------------------------------------------------------ | ------------------------ |
| rec20160318_191511_232p3_2cm_cont__4097im_1500ms_ML17keV_6.h5 | 232p3, wet |
| rec20160324_055424_232p1_wet_1cm_cont_4097im_1500ms_17keV_13_a.h5 | 232p1, wet |
| rec20160318_223946_244p1_1p5cm_cont__4097im_1500ms_ML17keV_7.h5 | 244p1, cured |
| rec20160324_123639_235p1_wet_0p7cm_cont_4097im_1500ms_17keV_14.h5 | 235p1, wet |
| rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5 | 244p1, wet |
| rec20160326_175540_235p4_wet_1p15cm_cont_4097im_1500ex_17keV_20.h5 | 235p4, wet |
| rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5_REG | 244p1, cured registered |
| rec20160327_003824_235p4_cured_1p15cm_cont_4097im_1500ex_17keV_22.h5 | 235p4, cured |
| rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5 | 232p3, cured |
| rec20160327_003824_235p4_cured_1p15cm_cont_4097im_1500ex_17keV_22.h5_REG | 235p4, cured registered |
| rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5_REG | 232p3, cured registered |
| rec20160327_160624_245p1_wet_1cm_cont_4097im_1500ex_17keV_23.h5 | 245p1, wet |
The first three numeric characters (e.g., 232 in 232p3) correspond to a material sample, and the last character (e.g., 3 in 232p3) correspond to different extrinsic factors, e.g. deformation. Despite being samples from similar materials, the reconstructed files presented several differences: different amount of ringing artifacts, intensity variation, noise, etc. For more info on the original data, please refer to [Larson et al (2019)](https://www.sciencedirect.com/science/article/abs/pii/S1359835X18304603).
The filenames for each processed sample follow the structure `-.zip`, where `` can be `tiramisu`, `tiramisu_3d`, `unet`, `unet_3d`. For example, results for the sample 232p3, wet, obtained with the 2D U-net network are given in the file `unet-rec20160318_191511_232p3_2cm_cont__4097im_1500ms_ML17keV_6.h5.zip`. This file contains the following structure, which is similar to the structure of the other ones containing predictions:
```bash
unet-rec20160318_191511_232p3_2cm_cont__4097im_1500ms_ML17keV_6.h5/
└── predict/
├── 000000.png
├── 000001.png
├── ...
└── 000999.png
```
Each `png` image in the folder `predict` correspond to a slice of the processed data.
## File `coefficients.zip`
This file contains:
* The training coefficients for each network.
* Filenames follow the structure `larson_.hdf5`.
* Accuracy, loss, validation accuracy and validation loss we obtained during our training process.
* Filenames follow the structure `larson_.hdf5-.csv`, where `` can be `accuracy`, `loss`, `val_accuracy`, `val_loss`, for accuracy, loss, validation accuracy and validation loss, respectively.
* Output of the training and prediction steps in our study.
* Filenames follow the structure `output.train_.txt` and `output.predict_.txt` for the training and prediction processes, respectively.
These file are contained in folders according to each network. The folder structure follows:
```bash
coefficients
├── tiramisu
│ ├── larson_tiramisu-67.hdf5
│ ├── larson_tiramisu-67.hdf5-accuracy.csv
│ ├── larson_tiramisu-67.hdf5-loss.csv
│ ├── larson_tiramisu-67.hdf5-val_accuracy.csv
│ ├── larson_tiramisu-67.hdf5-val_loss.csv
│ ├── output.predict_tiramisu-67.txt
│ └── output.train_tiramisu-67.txt
├── tiramisu_3d
│ ├── larson_tiramisu_3d-67.hdf5
│ ├── larson_tiramisu_3d-67.hdf5-accuracy.csv
│ ├── larson_tiramisu_3d-67.hdf5-loss.csv
│ ├── larson_tiramisu_3d-67.hdf5-val_accuracy.csv
│ ├── larson_tiramisu_3d-67.hdf5-val_loss.csv
│ ├── output.predict_tiramisu_3d-67.txt
│ └── output.train_tiramisu_3d-67.txt
├── unet
│ ├── larson_unet.hdf5
│ ├── larson_unet.hdf5-accuracy.csv
│ ├── larson_unet.hdf5-loss.csv
│ ├── larson_unet.hdf5-val_accuracy.csv
│ ├── larson_unet.hdf5-val_loss.csv
│ ├── output.predict_unet.txt
│ └── output.train_unet.txt
└── unet_3d
├── larson_unet_3d.hdf5
├── larson_unet_3d.hdf5-accuracy.csv
├── larson_unet_3d.hdf5-loss.csv
├── larson_unet_3d.hdf5-val_accuracy.csv
├── larson_unet_3d.hdf5-val_loss.csv
├── output.predict_unet_3d.txt
└── output.train_unet_3d.txt
4 directories, 28 files
```
## References
Larson, N. M., Cuellar, C. & Zok, F. W. [X-ray computed tomography of microstructure evolution during matrix impregnation and curing in unidirectional fiber beds.](https://www.sciencedirect.com/science/article/abs/pii/S1359835X18304603) Composites Part A: Applied Science and Manufacturing 117, 243–259 (2019).
# Appendix
Files included in this dataset:
```bash
README.md (this file)
coefficients.zip
tiramisu_3d-rec20160318_191511_232p3_2cm_cont__4097im_1500ms_ML17keV_6.h5.zip
tiramisu_3d-rec20160318_223946_244p1_1p5cm_cont__4097im_1500ms_ML17keV_7.h5.zip
tiramisu_3d-rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5_REG.zip
tiramisu_3d-rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5.zip
tiramisu_3d-rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5_REG.zip
tiramisu_3d-rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5.zip
tiramisu_3d-rec20160324_055424_232p1_wet_1cm_cont_4097im_1500ms_17keV_13_a.h5.zip
tiramisu_3d-rec20160324_123639_235p1_wet_0p7cm_cont_4097im_1500ms_17keV_14.h5.zip
tiramisu_3d-rec20160326_175540_235p4_wet_1p15cm_cont_4097im_1500ex_17keV_20.h5.zip
tiramisu_3d-rec20160327_003824_235p4_cured_1p15cm_cont_4097im_1500ex_17keV_22.h5_REG.zip
tiramisu_3d-rec20160327_003824_235p4_cured_1p15cm_cont_4097im_1500ex_17keV_22.h5.zip
tiramisu_3d-rec20160327_160624_245p1_wet_1cm_cont_4097im_1500ex_17keV_23.h5.zip
tiramisu-rec20160318_191511_232p3_2cm_cont__4097im_1500ms_ML17keV_6.h5.zip
tiramisu-rec20160318_223946_244p1_1p5cm_cont__4097im_1500ms_ML17keV_7.h5.zip
tiramisu-rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5_REG.zip
tiramisu-rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5.zip
tiramisu-rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5_REG.zip
tiramisu-rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5.zip
tiramisu-rec20160324_055424_232p1_wet_1cm_cont_4097im_1500ms_17keV_13_a.h5.zip
tiramisu-rec20160324_123639_235p1_wet_0p7cm_cont_4097im_1500ms_17keV_14.h5.zip
tiramisu-rec20160326_175540_235p4_wet_1p15cm_cont_4097im_1500ex_17keV_20.h5.zip
tiramisu-rec20160327_003824_235p4_cured_1p15cm_cont_4097im_1500ex_17keV_22.h5_REG.zip
tiramisu-rec20160327_003824_235p4_cured_1p15cm_cont_4097im_1500ex_17keV_22.h5.zip
tiramisu-rec20160327_160624_245p1_wet_1cm_cont_4097im_1500ex_17keV_23.h5.zip
unet_3d-rec20160318_191511_232p3_2cm_cont__4097im_1500ms_ML17keV_6.h5.zip
unet_3d-rec20160318_223946_244p1_1p5cm_cont__4097im_1500ms_ML17keV_7.h5.zip
unet_3d-rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5_REG.zip
unet_3d-rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5.zip
unet_3d-rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5_REG.zip
unet_3d-rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5.zip
unet_3d-rec20160324_055424_232p1_wet_1cm_cont_4097im_1500ms_17keV_13_a.h5.zip
unet_3d-rec20160324_123639_235p1_wet_0p7cm_cont_4097im_1500ms_17keV_14.h5.zip
unet_3d-rec20160326_175540_235p4_wet_1p15cm_cont_4097im_1500ex_17keV_20.h5.zip
unet_3d-rec20160327_003824_235p4_cured_1p15cm_cont_4097im_1500ex_17keV_22.h5_REG.zip
unet_3d-rec20160327_003824_235p4_cured_1p15cm_cont_4097im_1500ex_17keV_22.h5.zip
unet_3d-rec20160327_160624_245p1_wet_1cm_cont_4097im_1500ex_17keV_23.h5.zip
unet-rec20160318_191511_232p3_2cm_cont__4097im_1500ms_ML17keV_6.h5.zip
unet-rec20160318_223946_244p1_1p5cm_cont__4097im_1500ms_ML17keV_7.h5.zip
unet-rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5_REG.zip
unet-rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5.zip
unet-rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5_REG.zip
unet-rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5.zip
unet-rec20160324_055424_232p1_wet_1cm_cont_4097im_1500ms_17keV_13_a.h5.zip
unet-rec20160324_123639_235p1_wet_0p7cm_cont_4097im_1500ms_17keV_14.h5.zip
unet-rec20160326_175540_235p4_wet_1p15cm_cont_4097im_1500ex_17keV_20.h5.zip
unet-rec20160327_003824_235p4_cured_1p15cm_cont_4097im_1500ex_17keV_22.h5_REG.zip
unet-rec20160327_003824_235p4_cured_1p15cm_cont_4097im_1500ex_17keV_22.h5.zip
unet-rec20160327_160624_245p1_wet_1cm_cont_4097im_1500ex_17keV_23.h5.zip
```