Scripts for processing and converting raw data to label format and creating DCL projects
Currently, the script takes 3 inputs:
- .npz file that contains arrays for X (raw image) and y (segmentation mask) in the dimension order (TYXC)
- Data-registry metadata file
- Dataset configuration file
And outputs:
- .zip file in DeepCell Label format which has:
- cellTypes.json, which has each type in the "Cell Type Master List" but is otherwise empty
- X.ome.tiff, the raw data with the names of channels constituting intersection of the "Channel Master List" and the channels specified in the metadata.
- y.ome.tiff, the segmentation mask
To use the script, run:
python raw_to_dcl.py [--tile_x WIDTH] [--tile_y HEIGHT] /path/to/raw.npz /path/to/metadata.yaml.dvc /path/to/config.yaml output.zip]- Use dimension order in the metadata file to determine how to reorder X.ome.tiff
- Point to a dvc file instead of an npz file, pull that npz, and then create the project
- Allow user to programatically create a DCL project (ie. with a POST request)
- This may require changing how DCL creates projects since I think the POST request does not take zip files
- Integrate deepcell-types model to generate embeddings
- More command-line arguments and flexibility, for example:
- Including ground-truth in cellTypes.json
- Tiling options; tile sizes, whether to create separate projects or same file
- Combining n images into a single project
- Tests and exception handling