File size: 3,144 Bytes
26dea89
 
 
 
 
 
 
 
 
 
 
 
5f52fd6
26dea89
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5f52fd6
 
26dea89
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
{
  "name": "MitoEM (publicly-labeled half)",
  "challenge": "MitoEM \u2014 MICCAI 2020 / ISBI 2021",
  "papers": [
    "10.1007/978-3-030-59722-1_7",
    "10.1109/TMI.2023.3320497"
  ],
  "source": {
    "images": "https://huggingface.co/datasets/pytc/EM30",
    "labels": "https://huggingface.co/datasets/pytc/MitoEM",
    "challenge": "https://mitoem.grand-challenge.org/"
  },
  "license": "CC BY 4.0",
  "modality": "serial-section multi-beam SEM (ssSEM)",
  "resolution_nm_xyz": [
    8,
    8,
    30
  ],
  "body_part": "brain cortex (rat V1 L2/3; human L2)",
  "task": "mitochondria instance segmentation",
  "splits": {
    "train": 800,
    "val": 200
  },
  "ground_truth": "v2 instance labels (corrected release used by the IEEE TMI 2023 challenge report); uint16, 0=background, non-zero=instance ID; single tier, no competing raters",
  "excluded": [
    "z 500-999 of both volumes (challenge test half; GT withheld by organizers)"
  ],
  "subsets": {
    "MitoEM-H": {
      "organism": "human",
      "tissue": "cortex, Layer II",
      "resolution_nm_xyz": [
        8,
        8,
        30
      ],
      "n_instances_labeled": 10552,
      "source_images": "pytc/EM30 :: EM30-H-im-pad.zip",
      "source_labels": "pytc/MitoEM :: EM30-H-mito-train-val-v2.zip",
      "preprocessing": "cropped [0:4096, 0:4096] from the 5120x5120 padded slices"
    },
    "MitoEM-R": {
      "organism": "rat",
      "tissue": "primary visual cortex (V1), Layer II/III",
      "resolution_nm_xyz": [
        8,
        8,
        30
      ],
      "n_instances_labeled": 5446,
      "source_images": "pytc/EM30 :: EM30-R-im.zip",
      "source_labels": "pytc/MitoEM :: EM30-R-mito-train-val-v2.zip",
      "preprocessing": "none (already 4096x4096)"
    }
  },
  "loader_notes": [
    "binary semantic mitochondria = mask > 0",
    "instance IDs are volume-global and sparse; not contiguous within a slice",
    "both subsets are pre-aligned at 4096x4096; no crop/offset needed on read",
    "MitoEM-H upstream is 5120x5120 padded at the FAR edges -- the label frame is [0:4096, 0:4096], NOT the [512:4608] implied by 'pad-20-512-512'",
    "annotated instances have a minimum size of 2000 voxels",
    "MOAS and small/medium/large bins are evaluation strata, not label classes"
  ],
  "overlap_warning": "MitoEM-H and AxonEM-Human are the SAME image volume (EM30-H); AxonEM is already mirrored at MedOtter/AxonEM. 5 of its 9 human crops intersect MitoEM's labeled range. Targets differ (axons vs mitochondria) so this is benchmark non-independence, not label leakage. MitoEM-R is clean.",
  "license_note": "Upstream pytc/EM30 + pytc/MitoEM declare MIT, which covers the ANNOTATIONS. MitoEM-H is the EM30-H human volume whose governing imagery layer is the H01 release (Shapson-Coe et al.) under CC BY 4.0, so the mirror is tagged with the most restrictive governing layer -- matching MedOtter/AxonEM, served from the same pytc/EM30 archive. Both licences permit redistribution; only the attribution obligation differs. MitoEM-R's rat volume has no separately adjudicated upstream; CC BY 4.0 applied uniformly as the conservative choice."
}