Torch3Dseg - Dataset

Dataset for Training a 3D-UNet for semantic segmentation of textile reinforcements. The Following zip files and xlxs file have been uploaded: ├── 285-10_layer_00 │ ├── 285_10-layer_00_0 │ │ ├── 285_10-layer_00_0.seg.nrrd │ │ ├── 285_10-layer_00_0.vol.nrrd │ │ └── 285_10-layer_00_0.vol.seg.train.h5 │ ├── 285_10-layer_00_1 │ │ ├── 285_10-layer_00_1.seg.nrrd │ │ ├── 285_10-layer_00_1.vol.nrrd │ │ └── 285_10-layer_00_1.vol.seg.train.h5 │ └── 285_10-layer_00_3 │ ├── 285_10-layer_00_3.seg.nrrd │ ├── 285_10-layer_00_3.vol.nrrd │ └── 285_10-layer_00_3.vol.seg.eval.h5 ├── 285-10_layer_00.zip ├── 285-10_layer_01 │ ├── 285_10-layer_01_0 │ │ ├── 285_10-layer_01_0.seg.nrrd │ │ ├── 285_10-layer_01_0.vol.nrrd │ │ └── 285_10-layer_01_0.vol.seg.train.h5 │ ├── 285_10-layer_01_1 │ │ ├── 285_10-layer_01_1.seg.nrrd │ │ ├── 285_10-layer_01_1.vol.nrrd │ │ └── 285_10-layer_01_1.vol.seg.train.h5 │ └── 285_10-layer_01_3 │ ├── 285_10-layer_01_3.seg.nrrd │ ├── 285_10-layer_01_3.vol.nrrd │ └── 285_10-layer_01_3.vol.seg.eval.h5 ├── 285-10_layer_02.zip │ └── 285_10-layer_02_0 │ ├── 285_10-layer_02_0.seg.nrrd │ ├── 285_10-layer_02_0.vol.nrrd │ ├── 285_10-layer_02_0.vol.seg.eval.h5 │ ├── 285_10-layer_02_0.vol.seg.h5 │ └── 285_10-layer_02_0.vol.seg.train.h5 ├── 285-10_layer_03.zip │ └── 285_10-layer_03_0 │ ├── 285_10-layer_03_0.seg.nrrd │ ├── 285_10-layer_03_0.vol.nrrd │ ├── 285_10-layer_03_0.vol.seg.eval.h5 │ ├── 285_10-layer_03_0.vol.seg.h5 │ └── 285_10-layer_03_0.vol.seg.train.h5 └── 285_10_layer_machine_data.xlsx The zips contain the labeled data sets per stage. In each subfolder one can find the *.nrrd file of the volume and labeled segmentation. In the h5 file, you will find the following data: volume: Contains the raw X-ray computed tomography (CT) data, stored as a three-dimensional array of int16 values with shape [x_0, y_0, z_0]. This dataset represents the greyscale attenuation values in the scanned volume. labels: Holds the corresponding ground truth segmentation map, also as a three-dimensional array of int16 values with identical shape [x_0, y_0, z_0]. Each voxel is assigned a discrete label index representing the semantic class (e.g., background, weft, or fill yarn). masks: Provides binary masks as a one-hot channel for each semantic class, stored as a four-dimensional boolean array with shape [c, x_0, y_0, z_0], where c denotes the number of distinct classes. This dataset was used for training and evaluation. instances: Provides an instance label map for each yarn in weft and fill direction as a three-dimensional array [x_0, y_0, z_0] of uint16.