Skip to content

Latest commit

 

History

History

data

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
README.md
README.md
 
 
convert_to_zarr_h5.py
convert_to_zarr_h5.py
 
 
neuroglancer
neuroglancer
 
 
requirements.txt
requirements.txt
 
 

README.md

Download links and data prep scripts for the XPRESS datasets.

Downloading datasets

Install dependencies for the below scripts

pip3 install -r requirements.txt

Download image data

Training raw: download

Training voxel labels: download

Validation raw: download

Test raw: download

Download ground-truth skeletons

The skeletons are stored as NetworkX graphs in .npz files. Please see the ../eval/eval.py script to see how to properly read these skeleton files.

Training skeletons: download

Validation skeletons: download

Note: the GT skeletons are only fully annotated for the inner 33um volume for training and test, but is the 23.1um for validation. That is:

  • ROI of raw for all three volumes are 39.6 x 39.6 x 39.6 um (1200 x 1200 x 1200 pix at 33nm resolution).
  • ROI of GT skeletons for training/test is the inner 33 x 33 x 33 um (1000 x 1000 x 1000 pix with an offset of 100 x 100 x 100 pix).
  • ROI of GT skeletons for validation is the inner 23.1 x 23.1 x 23.1 um (700 x 700 x 700 pix with an offset of 250 x 250 x 250 pix).

(Optional) Convert h5 files to Zarr

For high performance training in the baseline model, we will need to convert .h5 files to chunked .zarr files.

python3 convert_to_zarr_h5.py xpress-training-raw.h5 volumes/raw xpress-challenge.zarr volumes/training_raw
python3 convert_to_zarr_h5.py xpress-training-voxel-labels.h5 volumes/labels xpress-challenge.zarr volumes/training_gt_labels
python3 convert_to_zarr_h5.py xpress-validation-raw.h5 volumes/raw xpress-challenge.zarr volumes/validation_raw
python3 convert_to_zarr_h5.py xpress-test-raw.h5 volumes/raw xpress-challenge.zarr volumes/test_raw

(Optional) Convert h5 files to image stack

If your workflow requires stack of 2D images, you'll need to extract them from the .h5 files. TODO: we may provide an example script later.

Visualize downloaded data

We'll use the neuroglancer script adapted from https://github.com/funkelab/funlib.show.neuroglancer/blob/master/scripts/neuroglancer to visualize the Zarr dataset.

Training dataset: ./neuroglancer --file xpress-challenge.zarr --datasets volumes/training_raw volumes/training_gt_labels

Validation dataset: ./neuroglancer --file xpress-challenge.zarr --datasets volumes/validation_raw

Test dataset: ./neuroglancer --file xpress-challenge.zarr --datasets volumes/test_raw