Where things come from
External references
The upstream segmenter (flood-filling networks)
The fragments this project joins come from an FFN pipeline; what it outputs, and what of that reaches us, is DATA.md §2d. The method itself:
- Flood-Filling Networks — Januszewski, Maitin-Shepard, Li, Kornfeld, Denk, Jain, arXiv 2016. The original: a recurrent 3D convolutional network that "directly produces individual segments from a raw image", replacing boundary detection plus clustering with one end-to-end object-at-a-time process.
- High-precision automated reconstruction of neurons with flood-filling networks — Januszewski, Kornfeld, Li, Pope, Blakely, Lindsey, Maitin-Shepard, Tyka, Denk, Jain, Nature Methods 2018. Mean error-free neurite path length 1.1 mm on zebra finch. Preprint.
- Automated Reconstruction of a Serial-Section EM Drosophila Brain with FFNs and Local Realignment — Li, Lindsey, Januszewski, Zheng, Bates et al., bioRxiv 2019.
- Improving Connectomics by an Order of Magnitude — the Google Research summary.
- Viren Jain's publications — all of the FFN work is electron microscopy; there is no light-sheet or expansion-microscopy paper, so the exaSPIM application is unpublished.
roi/flood.py runs our own ROI model the way an FFN runs, and docs/MODELS.md §5b says what that is and is not.
The imaging
- ExA-SPIM scientific overview — whole mouse brain expansion plus a custom large-format light-sheet microscope, 250 nm near-isotropic, no slicing.
- Brain-wide anatomy at cellular resolution — the platform page that names the FFN + GNN segmentation pipeline.
The atlas
- Allen Institute CCF 2017 annotation volume and structure ontology — see DATA.md §2c for the files, their sources
and their licence, and
exaproof.io.ccffor the registration this project builds from the publisheddisplacement_swcs.
Other
- https://mogen-release.web.app/