3D Nuclei annotations and StarDist 3D model(s) (rat brain)

<strong>Name</strong>: 3D Nuclei annotations and StarDist3D model(s) (rat brain) <strong><em>Images: </em></strong>From a large tiling acquisition ( https://doi.org/10.5281/zenodo.6646128 ) individual Tile (xyz : 1024x1024x62) were downsampled and cropped (128x128x62). Four crops, from different tiles (./annotations_BIOP/images/) were manually annotated with ITK-SNAP (./annotations_BIOP/masks/) These four images, and their corresponding masks, were cropped into four quadrants (./crops_BIOP_v1/) in order to get 16 different images (64x64x62). <strong><em>Conda environment</em></strong><em>: </em>A conda environment was created using the yml file <em>stardist0.8_TF1.15.yml</em> <strong><em>Training : </em></strong>Training was performed using the jupyter notebook <em>1-Training_notebook.ipynb</em>.<br> Three different trainings (with the same random seed, same anisotropy, patch size and grid) were performed and produced three different models (./models/) Validation images (from the random seed used) were exported to ease the visual inspection of the results(./val_rdm42/). <strong><em>Validation: </em></strong>To save metrics in a csv file and compare predictions to the annotations the jupyter notebook <em>2-QC_notebook.ipynb </em>can be used on the validation folder. <strong>Large images</strong>: To test the model on larger images one can use Whole_ds441.tif (or Crop_ds441.tif )<br> These images were obtained using the plugin BigSticher on the raw data ( https://doi.org/10.5281/zenodo.6646128 ), resaved as h5 and exported the downsample by 4 version.