Skip to content

Neuroglancer Precomputed

WEBKNOSSOS can read Neuroglancer precomputed datasets.

Neuroglancer Precomputed datasets can both be uploaded to WEBKNOSSOS through the web uploader or streamed from a remote server or the cloud.


You can try the Neuroglancer Precomputed support with the following datasets. Load them in WEBKNOSSOS as a remote dataset:

  • Mouse Cortex EM data hosted on Google Cloud Storage
    • gs://iarpa_microns/minnie/minnie65/em
    • Source: MICrONs Consortium et al. Functional connectomics spanning multiple areas of mouse visual cortex. bioRxiv 2021.07.28.454025; doi:
  • FlyEM Hemibrain hosted on Google Cloud Storage
  • Interphase HeLa cell EM data hosted on AWS S3
    • s3://janelia-cosem-datasets/jrc_hela-3/neuroglancer/em/fibsem-uint8.precomputed
    • Source: Open Organelle Project

Neuroglancer Precomputed folder structure

WEBKNOSSOS expects the following file structure for Neuroglancer Precomputed datasets:

my_dataset.precomputed             # One root folder per dataset
├─ info                            # Dataset [metadata in JSON format](
├─ scale_1                         # One subdirectory with the same name as each scale/magnification "key" value specified in the info file. Each subdirectory contains a chunked representation of the data for a single resolution.
│  ├─ <chunks>
│  └─ ...
├─ ...                  
└─ scale_n                  

For details see the Neuroglancer spec.

Performance Considerations

To get the best streaming performance for Neuroglancer Precomputed datasets consider the following settings.

  • Use chunk sizes of 32 - 128 voxels^3
  • Enable sharding