fidoptLucchi++, mitochondria of the mouse hippocampus
A block of a FIB-SEM volume of mouse CA1 hippocampus, 5 nm a voxel, its mitochondria segmented by hand (EPFL, re-annotated as Lucchi++).
8 test casesmouse hippocampusimage EMresearch only
Read before comparing (5)
- Source. The Lucchi++ archive (casser.io) is gone and EPFL's file server did not answer: images and labels are read from the Hugging Face copy of PyTorch Connectomics, from the Lucchi++ authors' laboratory, pinned by revision and MD5. Its images are EPFL's TIFFs voxel for voxel. Its test labels meet the Lucchi++ copy of EMPIAR-10982 at a Dice of 0.98, apart at the outlines, and EPFL's original labels at 0.88.
- Ranked apart. One mouse: the eight test blocks are parts of one sub-volume and correlated, not independent subjects.
- Overlap. MitoNetTEM, from EMPIAR-10982, leaves its FIB-SEM volumes out, Lucchi++ among them; no other dataset of the corpus holds the EPFL volume.
- Start. Mitochondria are 2.9 to 10.8% of a test block: the whole block named mitochondria scores a median Dice of 0.13, its dark voxels (Otsu) 0.24.
- Licence. Neither source names a licence: ask before any use beyond research.
No method has a result on every case of this dataset yet.
Case by case
Click a score to look at that result: the image, the truth in green and the method's map in red. Underlined scores have their image; for the others you see the case without the result.
About the data
- Test cases
- The provider's testing sub-volume taken whole, cut into eight blocks, halves along each axis, and the eight blocks of its training sub-volume for development. Both sub-volumes are neighbouring parts of one mouse's volume.
- Truth
- contours by experts Lucchi++, the mitochondria of the two EPFL sub-volumes re-annotated by Casser et al.'s experts for consistent mitochondrial membranes, one binary mask
- Source
- www.epfl.ch/labs/cvlab/data/data-em
sites.google.com/view/connectomics
arxiv.org/abs/1812.06024
huggingface.co/datasets/pytc/tutorial/tree/3798f68d1688f44008a792e32b98f3d649385b81/lucchi%2B%2B
doi.org/10.1109/TMI.2011.2171705 - Licence
- none stated: EPFL makes the volume 'publicly available' and the Lucchi++ page calls its data 'freely available', neither names terms
8 cases for tuning, never tested
train_z000_y000_x0000, train_z000_y000_x0512, train_z000_y384_x0000, train_z000_y384_x0512, train_z082_y000_x0000, train_z082_y000_x0512, train_z082_y384_x0000, train_z082_y384_x0512