Datasets:
image imagewidth (px) 640 640 | mask imagewidth (px) 640 640 | subject_id stringclasses 56
values | eye_id stringlengths 3 5 | session int64 1 9 | frame_idx int64 1 28 | filename stringlengths 38 42 | control_points stringlengths 25.5k 46.5k | image_id stringlengths 38 42 |
|---|---|---|---|---|---|---|---|---|
001 | 001 | 1 | 11 | 001_F_GA41_BW2905_PA44_DG11_PF0_D1_S01_11 | {
"version": "5.5.0",
"flags": {},
"shapes": [
{
"label": "1",
"points": [
[
218.4375,
167.1875
]
],
"group_id": null,
"description": null,
"shape_type": "point",
"flags": {},
"mask": null
},
{
"label": "2",
... | 001_F_GA41_BW2905_PA44_DG11_PF0_D1_S01_11 | ||
001 | 001 | 2 | 3 | 001_F_GA41_BW2905_PA47_DG11_PF0_D1_S02_3 | {
"version": "5.5.0",
"flags": {},
"shapes": [
{
"label": "1",
"points": [
[
307.8787878787879,
188.7878787878788
]
],
"group_id": null,
"description": "",
"shape_type": "point",
"flags": {},
"mask": null
},
{
"l... | 001_F_GA41_BW2905_PA47_DG11_PF0_D1_S02_3 | ||
001 | 001 | 3 | 8 | 001_F_GA41_BW2905_PA50_DG11_PF0_D1_S03_8 | {
"version": "5.5.0",
"flags": {},
"shapes": [
{
"label": "1",
"points": [
[
198.69822485207098,
172.54437869822482
]
],
"group_id": null,
"description": "",
"shape_type": "point",
"flags": {},
"mask": null
},
{
... | 001_F_GA41_BW2905_PA50_DG11_PF0_D1_S03_8 | ||
002 | 002 | 1 | 2 | 002_M_GA40_BW4040_PA41_DG11_PF0_D1_S01_2 | {
"version": "5.5.0",
"flags": {},
"shapes": [
{
"label": "1",
"points": [
[
347.1875,
222.1875
]
],
"group_id": null,
"description": "",
"shape_type": "point",
"flags": {},
"mask": null
},
{
"label": "2",
... | 002_M_GA40_BW4040_PA41_DG11_PF0_D1_S01_2 | ||
002 | 002 | 2 | 2 | 002_M_GA40_BW4040_PA43_DG11_PF0_D1_S02_2 | {
"version": "5.5.0",
"flags": {},
"shapes": [
{
"label": "1",
"points": [
[
386.06060606060606,
273.03030303030306
]
],
"group_id": null,
"description": "",
"shape_type": "point",
"flags": {},
"mask": null
},
{
... | 002_M_GA40_BW4040_PA43_DG11_PF0_D1_S02_2 | ||
002 | 002 | 3 | 10 | 002_M_GA40_BW4040_PA45_DG11_PF0_D1_S03_10 | "{\n \"version\": \"5.5.0\",\n \"flags\": {},\n \"shapes\": [\n {\n \"label\": \"1\",\n (...TRUNCATED) | 002_M_GA40_BW4040_PA45_DG11_PF0_D1_S03_10 | ||
002 | 002-1 | 1 | 6 | 002_M_GA40_BW4040_PA41_DG11_PF0_D1_S01_6 | "{\n \"version\": \"5.4.1\",\n \"flags\": {},\n \"shapes\": [\n {\n \"label\": \"1\",\n (...TRUNCATED) | 002_M_GA40_BW4040_PA41_DG11_PF0_D1_S01_6 | ||
002 | 002-1 | 2 | 12 | 002_M_GA40_BW4040_PA43_DG11_PF0_D1_S02_12 | "{\n \"version\": \"5.4.1\",\n \"flags\": {},\n \"shapes\": [\n {\n \"label\": \"1\",\n (...TRUNCATED) | 002_M_GA40_BW4040_PA43_DG11_PF0_D1_S02_12 | ||
002 | 002-1 | 3 | 6 | 002_M_GA40_BW4040_PA45_DG11_PF0_D1_S03_6 | "{\n \"version\": \"5.4.1\",\n \"flags\": {},\n \"shapes\": [\n {\n \"label\": \"1\",\n (...TRUNCATED) | 002_M_GA40_BW4040_PA45_DG11_PF0_D1_S03_6 | ||
003 | 003 | 1 | 4 | 003_F_GA41_BW3950_PA42_DG11_PF0_D1_S01_4 | "{\n \"version\": \"5.5.0\",\n \"flags\": {},\n \"shapes\": [\n {\n \"label\": \"1\",\n (...TRUNCATED) | 003_F_GA41_BW3950_PA42_DG11_PF0_D1_S01_4 |
End of preview. Expand in Data Studio
COph100
Comprehensive Ophthalmology fundus image dataset of 100 infant eyes (with Retinopathy of Prematurity), originally curated for retinal image registration.
Description
This is the segmentation-oriented re-packaging of COph100 for EasyMedSeg. Each sample is a fundus photograph (640x480 RGB, RetCam) paired with an automatically generated binary retinal-vessel mask. The mask was produced by Hu et al. using an SS-MAF model trained on FIVES (paper-reported 91.56% Dice on FIVES test set).
The 100 eyes come from infants screened for ROP; each eye was photographed across 2-9 examination sessions. The current parquet contains 324 (image, mask) sessions.
Columns
| Column | Type | Description |
|---|---|---|
image |
Image | Fundus photograph (640x480 RGB) |
mask |
Image | Binary retinal-vessel mask (480x640, 0=bg, 1=vessel) |
subject_id |
string | First three digits of filename, e.g. 001 |
eye_id |
string | Folder ID, e.g. 001 or 001-1 (-1 denotes the second eye of a subject) |
session |
int | Examination session index (parsed from S<NN> in filename) |
frame_idx |
int | Image index within session (last token of filename) |
filename |
string | Original filename stem |
control_points |
string | Original LabelMe-format JSON with 10 manual control-point pairs per image (kept for registration users) |
Source
- Paper: Hu, Y., Gong, M., Qiu, Z., et al. "COph100: A comprehensive fundus image registration dataset from infants constituting the RIDIRP database." Scientific Data 12, 99 (2025). DOI: 10.1038/s41597-025-04426-w
- Figshare (labels): https://doi.org/10.6084/m9.figshare.27061084
- Parent dataset (fundus JPGs): Timkovič et al., "Retinal Image Dataset of Infants and Retinopathy of Prematurity," Scientific Data 11, 814 (2024). DOI: 10.1038/s41597-024-03409-7
License
CC BY 4.0
Notes
- The vessel masks are model-generated, not human-labeled. Use them as soft ground truth.
- The
_jpg_Label.nii.gzand_overlay.pngfiles from the original release are dropped here: the NIfTI files are transposed copies ofmask, and the overlays are derivable fromimage + mask. - For registration evaluation, consume
control_points(the original LabelMe JSON).
- Downloads last month
- 196