Dataset versions & statistics#

Every MedVision annotation is produced by a versioned planner (1.0.0 through 1.4.0), and the exact set of benchmark samples depends on which planner version you load. This page collects the version guidance, the per-subtask index files, and the aggregate per-dataset statistics you need to know which numbers apply to your run.

For what a sample actually is (and why several instances of a target on one slice count once), see Dataset concepts. For the env var that toggles the sample filter, see Loading data.

Important

Leaderboard results use annotation v1.0.0. All published leaderboard numbers are computed on the v1.0.0 annotations, with ambiguous cases (multi-instance targets) removed during metric calculation as a workaround. For new studies we recommend the latest annotation version (currently v1.4.0), which is what MedVision_PLANNER_VERSION=latest resolves to.

Subtasks map to dataset subsets#

Each benchmark subtask corresponds to a subset of the MedVision dataset. The per-subtask sample sizes are tabulated for every dataset version 1.0.01.4.0 under dataset-info/:

These lists are generated by streaming each config from the dataset, so a subtask appears only if that pin can actually load it. They are therefore narrower than the aggregate counts below for two reasons the summaries do not model: MAMA-MIA and PI-CAI are unloadable at a 1.2.0 pin (their annotations were withdrawn and republished as 1.2.1), and the AFIDs / PDDCA / VerSe biometry plans are paused at every pin, so A/D covers only Ceph-Biometrics-400 and FeTA24.

Because the quantitative tasks require pixel→mm arithmetic, the distribution of pixel sizes (physical spacing) across subtasks is provided in pixel_sizes__ds_v1.0.0.

For the full description of the source datasets, modalities, anatomies, annotation types, and returned fields, see the Hugging Face dataset repo.

Per-dataset statistics (version-invariant)#

The modality / image / slice / segmentation statistics below are the same for every dataset version — only the benchmark-annotation counts (Box / T/L / A/D) depend on the planner version, and among those only T/L changes. Each row is version-invariant, but the set of rows is not: the table lists the 31 datasets present in 1.4.0, so its Total row is specific to that release. These numbers are computed from the local benchmark plans by script/misc/summarize_datasets.sh.

The Seg. annotations column counts segmentation-mask (MaskSize) annotations; these are tracked separately and are not part of the Box / T/L / A/D benchmark-annotation counts in the next section.

Per-dataset stats — modality, images, slices, segmentation

Dataset

Modality

3D Images

3D Masks

2D Slices

Seg. annotations

ACDC

MRI

300

300

43,962

94,160

AFIDs

MRI

72

0

432

0

AMOS22

CT, MRI

360

360

251,637

1,215,776

AbdomenAtlas1.0Mini

CT

5,195

5,195

3,778,805

13,770,398

AbdomenCT-1K

CT

1,000

1,000

711,155

1,549,325

BCV15

CT

60

60

34,472

125,870

BraTS24

MRI

10,632

3,033

2,019,118

3,767,594

CAMUS

ultrasound

1,000

1,000

670,964

1,341,433

Ceph-Biometrics-400

X Ray

400

0

7,600

0

CrossMoDA

MRI

105

105

14,115

16,623

DEEP-PSMA

PET

200

200

50,341

50,341

FLARE22

CT

50

50

34,235

152,954

FeTA24

MRI

80

80

35,776

153,599

HNTSMRG24

MRI

300

300

56,078

62,424

ISLES24

MRI

298

149

97,228

97,228

KiPA22

CT

70

70

29,494

74,690

KiTS23

CT

489

489

190,642

291,550

LIDC-IDRI

CT

1,013

1,013

73,976

73,976

LNQ2023

CT

120

120

25,331

25,331

MAMA-MIA

MRI

1,506

1,506

214,408

214,408

MSD

CT, MRI

3,225

1,741

791,706

1,438,472

MSWAL

CT

484

484

108,246

125,126

OAIZIB-CM

MRI

507

507

358,728

922,989

PDDCA

CT

48

48

15,003

37,119

PI-CAI

MRI

425

425

42,857

42,857

SKM-TEA

MRI

310

155

173,690

475,828

ToothFairy2

CT

480

480

397,531

2,131,223

TopCoW24

CT, MRI

250

250

87,953

251,901

TotalSegmentator

CT, MRI

1,844

1,844

1,091,563

16,979,575

VerSe

CT

325

325

208,402

874,896

autoPET-III

CT, PET

2,076

1,038

360,638

360,638

Total (31)

33,224

22,327

11,976,086

46,718,304

Benchmark annotations by version#

The three quantitative tasks — Box (detection), T/L (tumor/lesion size), and A/D (biometrics) — contribute the benchmark-annotation counts. Only T/L annotations have ever been regenerated — most recently in 1.4.0, which regenerates all 12 T/L datasets (see below) — while Box and A/D are byte-identical in every release. A total can therefore move for either of two reasons: a T/L regeneration, or a release adding datasets.

v1.4.0: regenerated T/L annotations#

v1.4.0 regenerates the Tumor-Lesion-Size annotations of all 12 T/L datasets. Clusters are now selected by a physical size floor in millimetres — max(2.0 mm, 2 × the coarser in-plane spacing) of the measured plane — instead of a raw pixel count, a gate that silently discarded rotated ellipses is removed, and the ellipse fit is guarded against degenerate results (the method is described under Dataset concepts). Published T/L landmarks grow from 75,840 to 3,801,540 (50×); every other task and every previously published annotation file is unchanged. Full details are in the v1.4.0 release note.

Landmarks across all three planes, previous published version → v1.4.0:

Dataset

From

Landmarks

Factor

DEEP-PSMA

1.2.0

753 → 156,439

208×

LNQ2023

1.2.0

238 → 37,898

159×

autoPET-III

1.1.1

3,118 → 488,375

157×

LIDC-IDRI

1.2.0

515 → 59,179

115×

MAMA-MIA

1.2.1

5,071 → 369,419

73×

PI-CAI

1.2.1

409 → 28,324

69×

BraTS24

1.1.1

26,198 → 1,645,550

63×

MSD

1.1.1

12,914 → 712,247

55×

HNTSMRG24

1.1.1

3,188 → 69,093

22×

MSWAL

1.3.0

12,260 → 145,136

12×

KiTS23

1.1.1

8,034 → 76,016

9.5×

KiPA22

1.1.1

3,142 → 13,864

4.4×

Total

75,840 → 3,801,540

50×

The loader still discards multi-cluster slices, so published sample counts grow by less than the raw landmark counts above.

Warning

The train/test split moved on six datasets. Case counts per split are unchanged, but which cases land on each side changes for HNTSMRG24 (47%), KiPA22 (43%), KiTS23 (43%), MSD (42%), autoPET-III (41%) and BraTS24 (41%) — their earlier splits were force-aligned to v1.0.0, and v1.4.0 is their first natural seeded split. Do not compare a v1.4.0 test-set metric against a pre-1.4.0 one on those six datasets, and re-derive any cached split.

Per-version annotation counts#

Each row below sums over the datasets its release contained — 31 for 1.3.0/1.4.0, 30 for 1.2.0, 22 for the others. 1.3.0 adds MSWAL without changing any annotation logic; 1.4.0 changes only the T/L counts, but changes them by a factor of 20:

Planner version

Single-instance (filtered)

Multi-instance (unfiltered)

1.4.0 (default)

25,725,431

50,529,833

1.3.0

24,806,967

46,804,120

1.2.0

24,738,696

46,666,781

1.1.1

24,279,534

45,338,754

1.1.0

24,292,466

45,354,786

1.0.0 (leaderboard)

24,276,501

45,314,742

Each per-dataset cell reads total (Box · T/L · A/D …), and the donut figures show the same split — the outer ring by dataset, the inner ring by task. The figures and counts are generated from the local benchmark plans by script/misc/summarize_datasets.sh (source counts also saved as dataset_summary_filtered.json / dataset_summary_raw.json under each dataset-info/datasets_summary_v<version>/).

MedVision v1.4.0 (default) — donut + annotation counts

T/L regenerated across all 12 T/L datasets — see v1.4.0: regenerated T/L annotations above for what changed and why. Box and A/D counts are carried over from 1.3.0 unmodified.

Single-instance (filtered)

v1.4.0 single-instance donut

Multi-instance (unfiltered)

v1.4.0 multi-instance donut

Dataset

Single-instance (Box / T/L / A/D)

Multi-instance (Box / T/L / A/D)

ACDC

14,271 (Box 14,271)

94,160 (Box 94,160)

AFIDs

432 (A/D 432)

432 (A/D 432)

AMOS22

666,532 (Box 666,532)

1,215,776 (Box 1,215,776)

AbdomenAtlas1.0Mini

9,748,290 (Box 9,748,290)

13,770,398 (Box 13,770,398)

AbdomenCT-1K

1,041,588 (Box 1,041,588)

1,549,325 (Box 1,549,325)

BCV15

68,543 (Box 68,543)

125,870 (Box 125,870)

BraTS24

1,542,863 (Box 1,115,524 · T/L 427,339)

5,413,144 (Box 3,767,594 · T/L 1,645,550)

CAMUS

951,370 (Box 951,370)

1,341,433 (Box 1,341,433)

Ceph-Biometrics-400

7,600 (A/D 7,600)

7,600 (A/D 7,600)

CrossMoDA

4,076 (Box 4,076)

16,623 (Box 16,623)

DEEP-PSMA

12,919 (Box 2,179 · T/L 10,740)

206,780 (Box 50,341 · T/L 156,439)

FLARE22

104,211 (Box 104,211)

152,954 (Box 152,954)

FeTA24

49,412 (Box 49,087 · A/D 325)

153,924 (Box 153,599 · A/D 325)

HNTSMRG24

75,949 (Box 32,029 · T/L 43,920)

131,517 (Box 62,424 · T/L 69,093)

ISLES24

9,774 (Box 9,774)

97,228 (Box 97,228)

KiPA22

50,708 (Box 37,647 · T/L 13,061)

88,554 (Box 74,690 · T/L 13,864)

KiTS23

176,716 (Box 114,491 · T/L 62,225)

367,566 (Box 291,550 · T/L 76,016)

LIDC-IDRI

53,867 (Box 10,379 · T/L 43,488)

133,155 (Box 73,976 · T/L 59,179)

LNQ2023

7,355 (Box 1,653 · T/L 5,702)

63,229 (Box 25,331 · T/L 37,898)

MAMA-MIA

158,492 (Box 67,767 · T/L 90,725)

583,827 (Box 214,408 · T/L 369,419)

MSD

393,215 (Box 277,451 · T/L 115,764)

2,150,719 (Box 1,438,472 · T/L 712,247)

MSWAL

138,692 (Box 60,106 · T/L 78,586)

270,262 (Box 125,126 · T/L 145,136)

OAIZIB-CM

648,048 (Box 648,048)

922,989 (Box 922,989)

PDDCA

15,208 (Box 15,076 · A/D 132)

37,251 (Box 37,119 · A/D 132)

PI-CAI

28,219 (Box 5,455 · T/L 22,764)

71,181 (Box 42,857 · T/L 28,324)

SKM-TEA

262,338 (Box 262,338)

475,828 (Box 475,828)

ToothFairy2

1,413,979 (Box 1,413,979)

2,131,223 (Box 2,131,223)

TopCoW24

41,829 (Box 41,829)

251,901 (Box 251,901)

TotalSegmentator

7,603,455 (Box 7,603,455)

16,979,575 (Box 16,979,575)

VerSe

351,811 (Box 350,311 · A/D 1,500)

876,396 (Box 874,896 · A/D 1,500)

autoPET-III

83,669 (Box 31,794 · T/L 51,875)

849,013 (Box 360,638 · T/L 488,375)

Total (31)

25,725,431

50,529,833

MedVision v1.3.0 — donut + annotation counts

Adds MSWAL (CT, 484 volumes). No annotation logic changed — every 1.2.0 count is carried over unmodified.

Single-instance (filtered)

v1.3.0 single-instance donut

Multi-instance (unfiltered)

v1.3.0 multi-instance donut

Dataset

Single-instance (Box / T/L / A/D)

Multi-instance (Box / T/L / A/D)

ACDC

14,271 (Box 14,271)

94,160 (Box 94,160)

AFIDs

432 (A/D 432)

432 (A/D 432)

AMOS22

666,532 (Box 666,532)

1,215,776 (Box 1,215,776)

AbdomenAtlas1.0Mini

9,748,290 (Box 9,748,290)

13,770,398 (Box 13,770,398)

AbdomenCT-1K

1,041,588 (Box 1,041,588)

1,549,325 (Box 1,549,325)

BCV15

68,543 (Box 68,543)

125,870 (Box 125,870)

BraTS24

1,131,404 (Box 1,115,524 · T/L 15,880)

3,793,777 (Box 3,767,594 · T/L 26,183)

CAMUS

951,370 (Box 951,370)

1,341,433 (Box 1,341,433)

Ceph-Biometrics-400

7,600 (A/D 7,600)

7,600 (A/D 7,600)

CrossMoDA

4,076 (Box 4,076)

16,623 (Box 16,623)

DEEP-PSMA

2,273 (Box 2,179 · T/L 94)

51,094 (Box 50,341 · T/L 753)

FLARE22

104,211 (Box 104,211)

152,954 (Box 152,954)

FeTA24

49,412 (Box 49,087 · A/D 325)

153,924 (Box 153,599 · A/D 325)

HNTSMRG24

34,301 (Box 32,029 · T/L 2,272)

65,612 (Box 62,424 · T/L 3,188)

ISLES24

9,774 (Box 9,774)

97,228 (Box 97,228)

KiPA22

40,724 (Box 37,647 · T/L 3,077)

77,832 (Box 74,690 · T/L 3,142)

KiTS23

121,539 (Box 114,491 · T/L 7,048)

299,584 (Box 291,550 · T/L 8,034)

LIDC-IDRI

10,796 (Box 10,379 · T/L 417)

74,493 (Box 73,976 · T/L 517)

LNQ2023

1,698 (Box 1,653 · T/L 45)

25,569 (Box 25,331 · T/L 238)

MAMA-MIA

71,007 (Box 67,767 · T/L 3,240)

219,479 (Box 214,408 · T/L 5,071)

MSD

283,577 (Box 277,451 · T/L 6,126)

1,451,386 (Box 1,438,472 · T/L 12,914)

MSWAL

68,360 (Box 60,106 · T/L 8,254)

137,386 (Box 125,126 · T/L 12,260)

OAIZIB-CM

648,048 (Box 648,048)

922,989 (Box 922,989)

PDDCA

15,208 (Box 15,076 · A/D 132)

37,251 (Box 37,119 · A/D 132)

PI-CAI

5,848 (Box 5,455 · T/L 393)

43,266 (Box 42,857 · T/L 409)

SKM-TEA

262,338 (Box 262,338)

475,828 (Box 475,828)

ToothFairy2

1,413,979 (Box 1,413,979)

2,131,223 (Box 2,131,223)

TopCoW24

41,829 (Box 41,829)

251,901 (Box 251,901)

TotalSegmentator

7,603,455 (Box 7,603,455)

16,979,575 (Box 16,979,575)

VerSe

351,811 (Box 350,311 · A/D 1,500)

876,396 (Box 874,896 · A/D 1,500)

autoPET-III

32,673 (Box 31,794 · T/L 879)

363,756 (Box 360,638 · T/L 3,118)

Total (31)

24,806,967

46,804,120

MedVision v1.2.0 — donut + annotation counts

Single-instance (filtered)

v1.2.0 single-instance donut

Multi-instance (unfiltered)

v1.2.0 multi-instance donut

Dataset

Single-instance (Box / T/L / A/D)

Multi-instance (Box / T/L / A/D)

ACDC

14,271 (Box 14,271)

94,160 (Box 94,160)

AFIDs

432 (A/D 432)

432 (A/D 432)

AMOS22

666,532 (Box 666,532)

1,215,776 (Box 1,215,776)

AbdomenAtlas1.0Mini

9,748,290 (Box 9,748,290)

13,770,398 (Box 13,770,398)

AbdomenCT-1K

1,041,588 (Box 1,041,588)

1,549,325 (Box 1,549,325)

BCV15

68,543 (Box 68,543)

125,870 (Box 125,870)

BraTS24

1,131,404 (Box 1,115,524 · T/L 15,880)

3,793,777 (Box 3,767,594 · T/L 26,183)

CAMUS

951,370 (Box 951,370)

1,341,433 (Box 1,341,433)

Ceph-Biometrics-400

7,600 (A/D 7,600)

7,600 (A/D 7,600)

CrossMoDA

4,076 (Box 4,076)

16,623 (Box 16,623)

DEEP-PSMA

2,273 (Box 2,179 · T/L 94)

51,094 (Box 50,341 · T/L 753)

FLARE22

104,211 (Box 104,211)

152,954 (Box 152,954)

FeTA24

49,412 (Box 49,087 · A/D 325)

153,924 (Box 153,599 · A/D 325)

HNTSMRG24

34,301 (Box 32,029 · T/L 2,272)

65,612 (Box 62,424 · T/L 3,188)

ISLES24

9,774 (Box 9,774)

97,228 (Box 97,228)

KiPA22

40,724 (Box 37,647 · T/L 3,077)

77,832 (Box 74,690 · T/L 3,142)

KiTS23

121,539 (Box 114,491 · T/L 7,048)

299,584 (Box 291,550 · T/L 8,034)

LIDC-IDRI

10,796 (Box 10,379 · T/L 417)

74,493 (Box 73,976 · T/L 517)

LNQ2023

1,698 (Box 1,653 · T/L 45)

25,569 (Box 25,331 · T/L 238)

MAMA-MIA

71,094 (Box 67,767 · T/L 3,327)

219,525 (Box 214,408 · T/L 5,117)

MSD

283,577 (Box 277,451 · T/L 6,126)

1,451,386 (Box 1,438,472 · T/L 12,914)

OAIZIB-CM

648,048 (Box 648,048)

922,989 (Box 922,989)

PDDCA

15,208 (Box 15,076 · A/D 132)

37,251 (Box 37,119 · A/D 132)

PI-CAI

5,850 (Box 5,455 · T/L 395)

43,267 (Box 42,857 · T/L 410)

SKM-TEA

262,338 (Box 262,338)

475,828 (Box 475,828)

ToothFairy2

1,413,979 (Box 1,413,979)

2,131,223 (Box 2,131,223)

TopCoW24

41,829 (Box 41,829)

251,901 (Box 251,901)

TotalSegmentator

7,603,455 (Box 7,603,455)

16,979,575 (Box 16,979,575)

VerSe

351,811 (Box 350,311 · A/D 1,500)

876,396 (Box 874,896 · A/D 1,500)

autoPET-III

32,673 (Box 31,794 · T/L 879)

363,756 (Box 360,638 · T/L 3,118)

Total (30)

24,738,696

46,666,781

MedVision v1.1.1 — donut + annotation counts

Single-instance (filtered)

v1.1.1 single-instance donut

Multi-instance (unfiltered)

v1.1.1 multi-instance donut

Dataset

Single-instance (Box / T/L / A/D)

Multi-instance (Box / T/L / A/D)

ACDC

14,271 (Box 14,271)

94,160 (Box 94,160)

AMOS22

666,532 (Box 666,532)

1,215,776 (Box 1,215,776)

AbdomenAtlas1.0Mini

9,748,290 (Box 9,748,290)

13,770,398 (Box 13,770,398)

AbdomenCT-1K

1,041,588 (Box 1,041,588)

1,549,325 (Box 1,549,325)

BCV15

68,543 (Box 68,543)

125,870 (Box 125,870)

BraTS24

1,131,404 (Box 1,115,524 · T/L 15,880)

3,793,777 (Box 3,767,594 · T/L 26,183)

CAMUS

951,370 (Box 951,370)

1,341,433 (Box 1,341,433)

Ceph-Biometrics-400

7,600 (A/D 7,600)

7,600 (A/D 7,600)

CrossMoDA

4,076 (Box 4,076)

16,623 (Box 16,623)

FLARE22

104,211 (Box 104,211)

152,954 (Box 152,954)

FeTA24

49,412 (Box 49,087 · A/D 325)

153,924 (Box 153,599 · A/D 325)

HNTSMRG24

34,301 (Box 32,029 · T/L 2,272)

65,612 (Box 62,424 · T/L 3,188)

ISLES24

9,774 (Box 9,774)

97,228 (Box 97,228)

KiPA22

40,724 (Box 37,647 · T/L 3,077)

77,832 (Box 74,690 · T/L 3,142)

KiTS23

121,539 (Box 114,491 · T/L 7,048)

299,584 (Box 291,550 · T/L 8,034)

MSD

283,577 (Box 277,451 · T/L 6,126)

1,451,386 (Box 1,438,472 · T/L 12,914)

OAIZIB-CM

648,048 (Box 648,048)

922,989 (Box 922,989)

SKM-TEA

262,338 (Box 262,338)

475,828 (Box 475,828)

ToothFairy2

1,413,979 (Box 1,413,979)

2,131,223 (Box 2,131,223)

TopCoW24

41,829 (Box 41,829)

251,901 (Box 251,901)

TotalSegmentator

7,603,455 (Box 7,603,455)

16,979,575 (Box 16,979,575)

autoPET-III

32,673 (Box 31,794 · T/L 879)

363,756 (Box 360,638 · T/L 3,118)

Total (22)

24,279,534

45,338,754

MedVision v1.1.0 — donut + annotation counts

Single-instance (filtered)

v1.1.0 single-instance donut

Multi-instance (unfiltered)

v1.1.0 multi-instance donut

Dataset

Single-instance (Box / T/L / A/D)

Multi-instance (Box / T/L / A/D)

ACDC

14,271 (Box 14,271)

94,160 (Box 94,160)

AMOS22

666,532 (Box 666,532)

1,215,776 (Box 1,215,776)

AbdomenAtlas1.0Mini

9,748,290 (Box 9,748,290)

13,770,398 (Box 13,770,398)

AbdomenCT-1K

1,041,588 (Box 1,041,588)

1,549,325 (Box 1,549,325)

BCV15

68,543 (Box 68,543)

125,870 (Box 125,870)

BraTS24

1,134,663 (Box 1,115,524 · T/L 19,139)

3,797,951 (Box 3,767,594 · T/L 30,357)

CAMUS

951,370 (Box 951,370)

1,341,433 (Box 1,341,433)

Ceph-Biometrics-400

7,600 (A/D 7,600)

7,600 (A/D 7,600)

CrossMoDA

4,076 (Box 4,076)

16,623 (Box 16,623)

FLARE22

104,211 (Box 104,211)

152,954 (Box 152,954)

FeTA24

49,412 (Box 49,087 · A/D 325)

153,924 (Box 153,599 · A/D 325)

HNTSMRG24

35,158 (Box 32,029 · T/L 3,129)

66,899 (Box 62,424 · T/L 4,475)

ISLES24

9,774 (Box 9,774)

97,228 (Box 97,228)

KiPA22

40,724 (Box 37,647 · T/L 3,077)

77,832 (Box 74,690 · T/L 3,142)

KiTS23

126,962 (Box 114,491 · T/L 12,471)

305,698 (Box 291,550 · T/L 14,148)

MSD

286,603 (Box 277,451 · T/L 9,152)

1,455,092 (Box 1,438,472 · T/L 16,620)

OAIZIB-CM

648,048 (Box 648,048)

922,989 (Box 922,989)

SKM-TEA

262,338 (Box 262,338)

475,828 (Box 475,828)

ToothFairy2

1,413,979 (Box 1,413,979)

2,131,223 (Box 2,131,223)

TopCoW24

41,829 (Box 41,829)

251,901 (Box 251,901)

TotalSegmentator

7,603,455 (Box 7,603,455)

16,979,575 (Box 16,979,575)

autoPET-III

33,040 (Box 31,794 · T/L 1,246)

364,507 (Box 360,638 · T/L 3,869)

Total (22)

24,292,466

45,354,786

MedVision v1.0.0 (leaderboard) — donut + annotation counts

Single-instance (filtered)

v1.0.0 single-instance donut

Multi-instance (unfiltered)

v1.0.0 multi-instance donut

Dataset

Single-instance (Box / T/L / A/D)

Multi-instance (Box / T/L / A/D)

ACDC

14,271 (Box 14,271)

94,160 (Box 94,160)

AMOS22

666,532 (Box 666,532)

1,215,776 (Box 1,215,776)

AbdomenAtlas1.0Mini

9,748,290 (Box 9,748,290)

13,770,398 (Box 13,770,398)

AbdomenCT-1K

1,041,588 (Box 1,041,588)

1,549,325 (Box 1,549,325)

BCV15

68,543 (Box 68,543)

125,870 (Box 125,870)

BraTS24

1,126,595 (Box 1,115,524 · T/L 11,071)

3,778,687 (Box 3,767,594 · T/L 11,093)

CAMUS

951,370 (Box 951,370)

1,341,433 (Box 1,341,433)

Ceph-Biometrics-400

7,600 (A/D 7,600)

7,600 (A/D 7,600)

CrossMoDA

4,076 (Box 4,076)

16,623 (Box 16,623)

FLARE22

104,211 (Box 104,211)

152,954 (Box 152,954)

FeTA24

49,412 (Box 49,087 · A/D 325)

153,924 (Box 153,599 · A/D 325)

HNTSMRG24

33,421 (Box 32,029 · T/L 1,392)

63,840 (Box 62,424 · T/L 1,416)

ISLES24

9,774 (Box 9,774)

97,228 (Box 97,228)

KiPA22

40,742 (Box 37,647 · T/L 3,095)

77,785 (Box 74,690 · T/L 3,095)

KiTS23

122,975 (Box 114,491 · T/L 8,484)

300,090 (Box 291,550 · T/L 8,540)

MSD

284,923 (Box 277,451 · T/L 7,472)

1,446,146 (Box 1,438,472 · T/L 7,674)

OAIZIB-CM

648,048 (Box 648,048)

922,989 (Box 922,989)

SKM-TEA

262,338 (Box 262,338)

475,828 (Box 475,828)

ToothFairy2

1,413,979 (Box 1,413,979)

2,131,223 (Box 2,131,223)

TopCoW24

41,829 (Box 41,829)

251,901 (Box 251,901)

TotalSegmentator

7,603,455 (Box 7,603,455)

16,979,575 (Box 16,979,575)

autoPET-III

32,529 (Box 31,794 · T/L 735)

361,387 (Box 360,638 · T/L 749)

Total (22)

24,276,501

45,314,742

Warning

Multi-instance annotations are not for leaderboard comparison. Do not use them to compare models on the leaderboard — the current MedVision-V0 SFT/RFT training is not optimized for multi-instance detection and measurement tasks.