# Scroll 1 Grand Prize Banner Region Dataset

This folder contains data related to the Grand Prize (GP) region from `Scroll1`. The dataset includes volumes and meshes associated to manual segmentation of surfaces.

## Contents

### 1. `volumes/`
This directory contains the volumetric data for the GP region.

- **`gp_tifstack.7z`**: 
  - Archive of the original uint16 TIFF stacks representing the volumetric data in the GP region.
  - The images have the original shape, but the non GP region is masked.
  - Data is not cropped to maintain alignment integrity, coordinate system is ZYX.

- **`gp_volume.zarr/`**: 
  - Multiresolution, standardized OME-Zarr containing the GP region from `Scroll1`.
  - The script to generate the standardized zarr is shared [here](https://github.com/ScrollPrize/villa/tree/main/foundation/scroll2zarr).

### 2. `gp_meshes.7z`
Archive containing all `.obj` triangular meshes for the GP region segments, semi-manually annotated with [Volume Cartographer](https://github.com/educelab/volume-cartographer). The coordinate system, as usual when dealing with meshes, is XYZ.

### 3. `predictions`
- **`gp_legendary-medial-cubes.7z`:
  - Archive containing a `.zarr` with segmentation medial surface predictions obtained with a multiclass semantic segmentation [nnUNet](https://dl.ash2txt.org/ml-models/segmentation/nnUNets/results/Dataset004_legendary-medial-cubes/nnUNetTrainer__nnUNetResEncUNetPlans_80G__3d_fullres_bs16/fold_0/checkpoint_best.pth).
  - Background is 0, medial surface is 5, labels from 4 to 1 are surface layers from the medial surface towards the border (1).
- **`gp_legendary-medial-cubes-softmax.7z`:
  - Output from the the same model as `gp_legendary-medial-cubes.7z`, but instead of taking the most likely label as final prediction, we provide the normalized average of each label weighted by its estimated probability.