OphGen — curated public and institutional ophthalmic datasets

Curated copies of 88 public and institutional ophthalmic imaging datasets, prepared for the OphGen ophthalmic generative-model project: images as stored in the OphGen image store (de-identified, de-duplicated, native resolution), the train / validation / test split used in the project, six captions per image (short / medium / long × English / Chinese) and all original task annotations.

Public datasets: use and licence follow each original dataset (see the dataset cards). Please cite the original publications and respect their terms.

Institutional datasets (private hospital data): access is restricted to the OphGen project; do not redistribute or re-upload. Images were de-identified (burned-in personal text removed); their source_paths are paths in the hospital archives.

Loading

from datasets import load_dataset
ds = load_dataset("Kaphathy/OphGen", "REFUGE2")            # one dataset, its own splits and labels
side = load_dataset("Kaphathy/OphGen", "REFUGE2-sidecar")  # its other annotation files (path, size, bytes), if any
pre = load_dataset("Kaphathy/OphGen", "all")               # every dataset (default config), e.g. for pretraining

In all, an image released by several datasets appears once per dataset (same image_id, same split); deduplicate on image_id if needed.

Datasets

Config	Dataset	Type	Modalities	Train	Validation	Test	Tasks
ADAM	ADAM (iChallenge-AMD official training set)	public	CFP	286	66	45	1
AGAR300	AGAR300	public	CFP	25	2	1	0
AIChallenger	AI Challenger 2018 retinal edema	public	OCT	8,960	1,920	1,920	1
AIDK	AIDK	public	AS-OCT	818	116	234	1
AMD-SD	AMD-SD	public	OCT	2,121	296	613	5
AROI	AROI	public	OCT	768	137	231	5
ASOCT-MSDC	ASOCT-MSDC	public	AS-OCT	773	110	221	5
Birdshot-Wide	Birdshot-Wide	public	UWF	3,865	553	1,104	2
BUltrasound	BUltrasound	institutional	Conventional ocular B-scan ultrasonography	101,473	5,689	5,574	0
CHAKSU	Chákṣu	public	CFP	903	100	335	3
chengdu_ffa	chengdu_ffa	institutional	FFA	3,519	473	934	1
ChengduShiyiUBM	ChengduShiyiUBM	institutional	UBM	413,564	22,972	22,970	0
CSDI	CSDI	public	CFP	131	19	37	4
CTI-FFA	CTI (FFA part)	public	FFA	1,244	138	174	3
CTI-ICGA	CTI (ICGA part)	public	ICGA	1,154	129	174	3
Data_on_OCT_and_Fundus	AFIO	public	OCT, CFP	70	12	16	4
DDR	DDR	public	CFP	6,552	2,688	4,232	3
DeepDRiD	DeepDRiD	public	CFP, UWF	1,329	446	450	3
DMR	DMR (Davis grading)	public	CFP	6,960	992	1,987	3
DRIVE	DRIVE	public	CFP	18	2	20	1
Duke_DME	Duke DME (Chiu)	public	OCT	77	11	22	3
Duke_iAMD	Duke iAMD	public	OCT	26,840	3,799	7,662	3
Duke_Srinivasan	Duke OCT (Srinivasan)	public	OCT	2,263	299	669	2
eryuanFundusT17RDCls	eryuanFundusT17RDCls	institutional	CFP	13,115	1,906	3,719	1
eyFRDCls	eyFRDCls	institutional	CFP	418	53	139	1
eyOCTset	eyOCTset	institutional	OCT	160,167	8,897	8,899	0
FairSeg	Harvard-FairSeg	public	IR/NIR	7,200	800	2,000	2
Fang_Denoising	Duke OCT denoising (Fang)	public	OCT	24	4	6	2
Final_eryuan_FFA	Final_eryuan_FFA	institutional	FFA	150,098	8,374	8,562	0
FIVES	FIVES	public	CFP	536	59	199	2
FQ_Datasets	FQ_Datasets	institutional	CFP	2,681	402	802	1
FUND	FUND	public	OCT, CFP	1,173	277	293	3
G1020	G1020	public	CFP	713	102	204	2
GAVE2	GAVE2	public	FFA, CFP	150	150	0	4
GLEAM	GLEAM	public	IR/NIR, OCT	1,436	480	480	1
handan_ffa	handan_ffa	institutional	FFA, CFP	26,560	1,478	1,471	0
handan_ffa2	handan_ffa2	institutional	FFA	17,465	980	991	0
HEI-MED	HEI-MED	public	CFP	118	17	34	6
HYAMD	HYAMD	public	CFP	1,084	164	312	3
HYGD	HYGD	public	CFP	515	75	147	2
IDRiD	IDRiD (segmentation and grading parts)	public	CFP	382	42	107	2
IOVS2015	IOVS2015 (Duke DME leakage)	public	FFA	401	57	115	3
JRC-FA_FAF_IR	JRC	public	FFA, FAF, IR/NIR	84	12	24	1
JSIEC	JSIEC	public	CFP	697	100	199	1
KAP_vit	KAP_vit	institutional	CFP	234,828	23,034	39,425	1
Labeled_Retinal_OCT	NEH OCT	public	OCT	11,686	1,665	3,334	4
MCOA	MCOA	public	AS-OCT, Slit-lamp anterior-segment photography	4,627	646	1,326	6
MFIDDR	MFIDDR	public	CFP	23,264	2,584	8,604	4
MMAC	MMAC 2023 (task 1)	public	CFP	1,136	247	0	3
MMRDR	MMRDR	public	CFP, UWF, OCT	17,075	1,897	5,380	4
OCT-C8	OCT-C8	public	OCT	6,746	1,051	1,034	1
OCT_MS_and_HC	JHU OCT	public	OCT	1,176	196	343	5
OCTA-25K	OCTA-25K	public	OCTA	12,505	695	695	0
OCTA500	OCTA-500	public	OCTA, OCT	277,645	14,140	70,700	4
OCTAVE	OCTAVE	public	OCT	2,584	361	741	4
OCTDL	OCTDL	public	OCT	1,406	201	402	4
OCTID	OCTID	public	OCT	399	60	113	2
Oculo	Oculo	public	Conventional ocular B-scan ultrasonography	1,189	162	245	2
ODIR	ODIR-5K	public	CFP	6,215	701	1,001	4
OIMHS	OIMHS	public	OCT	2,697	389	773	5
OLIVES	OLIVES	public	OCT, IR/NIR	104,590	14,305	29,361	8
ORIGA	ORIGA	public	CFP	293	32	325	4
OUWFD	OUWFD	public	UWF	490	70	140	5
PALM	PALM	public	CFP	386	401	397	3
PAPILA	PAPILA	public	CFP	342	48	98	6
PKU37	PKU37	public	OCT	1,211	147	370	2
PRIME-FP20	PRIME-FP20	public	FFA, UWF	20	4	6	3
Rare_Retinal_OCT	Rare retinal disease OCT images	public	OCT	134	19	38	2
RAVIR	RAVIR	public	IR/NIR	20	2	19	2
REFUGE2	REFUGE2	public	CFP	395	402	403	3
RETOUCH	RETOUCH	public	OCT	6,115	693	4,270	5
RFMiD	RFMiD 1.0	public	CFP	1,920	640	640	2
RLOCT	RLOCT (Reality Labs whole-eye OCT, anterior segment)	public	AS-OCT	34,123	4,558	10,079	4
ROSSA	ROSSA	public	OCTA	100	100	100	1
RVO-ME	RVO-ME	public	OCT	2,057	229	580	3
SJTU_Peripapillary	SJTU peripapillary OCT	public	OCT	74	24	24	3
SLID	SLID	public	Slit-lamp anterior-segment photography	2,355	131	131	0
tongren16G95DFundus	tongren16G95DFundus	institutional	CFP	172,345	24,419	49,171	1
tongren_archives	tongren_archives	institutional	CFP, External ocular photography	674,175	58,101	91,359	0
tongrenHandan	tongrenHandan	institutional	CFP	7,012	987	1,974	1
trhd3_FRD_Cls66_Train	trhd3_FRD_Cls66_Train	institutional	CFP	7,780	1,106	2,176	1
UESTC_OCT	UESTC OCT-R1/R2	public	OCT	29,184	3,072	2,000	1
UMN_DME	UMN DME	public	OCT	498	75	150	4
UWF-DR_2026	UWF-DR	public	UWF	1,135	158	327	2
UWF-Tumors	UWF intraocular tumours	public	UWF	1,410	201	403	1
yinhai_ffa	yinhai_ffa	institutional	FFA	812	34	52	0
yinhai_uwf	yinhai_uwf	institutional	UWF	175,209	9,684	9,790	0
YinhaiUBM	YinhaiUBM	institutional	UBM	75,720	4,012	3,857	0
Total				2,873,813	237,081	420,684

Split rules: official = official split; official_val_from_train = 10% of the official training set (by patient) used as validation; labeled_7_1_2 = patient-level 7:1:2 split with a fixed seed (no official split); pretrain_90_5_5 = 90:5:5 split for images without task labels. An image with different splits across tasks takes the strictest (test > validation > train).

Columns (dataset configs and all)

Column	Description
image_id	SHA-256 of the original file (stable identifier)
image	image bytes in the stored format; native resolution (no resizing)
split, split_rule	partition used in the project and how it was assigned
modality, modality_name	one of 13 standard ophthalmic modalities
width, height, format, field_of_view, device, eye, sex, age	metadata (empty when not documented)
representation, octa_layer, scan_area_mm	OCTA-specific fields
short_en … long_zh	captions built only from recorded facts; images in official test sets carry no task labels in captions (benchmark_holdout)
source_paths	original path(s) of the image in the official release (public) or hospital archive (institutional)
labels	task labels of this dataset: task, name (label name, value encoding included), value, num, official_split, annotation / label_file (path of the annotation, embedded in masks or stored in the -sidecar config)
masks	raster annotation masks for this image (name matches labels.name, path matches labels.annotation)

<dataset>-sidecar configs: path, size, bytes of every other annotation file of the release (tables, JSON, NPZ, MAT, XML …); write bytes to path to restore the original layout.

Processing

Exact and pixel-identical duplicates were merged (labels kept); images with burned-in personal text or non-removable overlays, undecodable or blank files were removed; images were not resized. Where the release did not document a mask value mapping, the mapping was inferred and is stated in the label name. Full methods: OphGen supplementary methods.