60 docs tagged with "manual"

The first public AGAR300 release contains 28 color fundus images with microaneurysms, captured at a 45 degree field of view and 2448 by 3264 pixel resolution.

De-identified fluorescein and indocyanine-green angiography images paired with structured lesion descriptions and reports.

Expert-curated ophthalmology multiple-choice benchmark with rationales, assembled for held-out language-model evaluation.

Longitudinal widefield fundus photographs from birdshot chorioretinitis eyes and age- and sex-matched controls.

Semantic segmentation labels for 4,670 frames from 25 cataract surgery videos (CATARACTS challenge). 25 anatomy and instrument classes.

Six near-infrared or synthetic iris subsets spanning close-range, lamp variation, twins, distance, large-scale, and synthetic recognition.

101 cataract surgery videos with 10-phase workflow annotations. Canonical older cataract benchmark.

1000 cataract surgery videos from multiple surgeons annotated for 10 surgical phases, instrument segmentation, and tool presence detection. First large-scale cataract surgical video dataset.

50 cataract surgery videos (>9 hours total) with frame-level annotations of 21 surgical tools. Earlier and larger sibling to Cataract-1K.

50 annotated 3D SD-OCT volumes (7,050 B-scans, 640×385 px) from subjects with vitreomacular adhesion (VMA, 25 eyes) and vitreomacular traction (VMT, 25 eyes). Each volume: 141 B-scans over 2×7×7 mm. A

28 fundus images with manual vessel segmentations (two annotators).

Professional/extended CORN database with in-vivo confocal microscopy images of the corneal subbasal nerve plexus. Supports nerve fiber segmentation, image quality enhancement, and tortuosity grading (

1,500 in-vivo confocal microscopy images of corneal subbasal nerve plexus, graded for tortuosity (4 levels). Train: 1,200 / Test: 300. 384×384 px, 400×400 µm FOV.

Combined corneal confocal microscopy collection comprising CORN-1, CORN-2, CORN-3, CORN-1500, CORN-Pro, and CORN-Complex. The subsets support corneal-nerve and cell segmentation, image enhancement, to

15,709 fundus images with paired medical reports and extracted keywords. Only public fundus report-generation dataset — useful for VLM / captioning evaluation.

35,126 fundus images organised for 5-class DR severity grading (ICDR grades 0-4): No DR 25,810 / Mild 2,443 / Moderate 5,292 / Severe 873 / Proliferative 708. Hosted on Alibaba Tianchi.

Fifty color fundus images with expert markings of diabetic retinopathy lesions, blood vessels, optic disc, and macula.

Forty color fundus photographs from a diabetic retinopathy screening program, divided into 20 training and 20 test images, with vessel reference annotations and field of view masks.

3,100 paired two-field fundus images with DR grading labels. Only public two-field paired DR benchmark.

Twenty OCT volumes with 220 selected B-scans from eyes with non-neovascular AMD, drusen, and geographic atrophy, including manual and automated pathology markings.

Paired low signal and high signal retinal OCT data used to study sparse acquisition, denoising, interpolation, and reconstruction in normal and non-neovascular AMD eyes.

OCT data from 384 subjects, including 269 with AMD and 115 normal subjects, with 38,400 B-scans and derived total retina and RPE-drusen complex thickness measurements.

Forty-five retinal OCT volumes acquired with a Spectralis system: 15 dry AMD, 15 diabetic macular edema, and 15 normal volumes. The official Duke release provides the study data.

Pediatric external-eye photographs with gaze-view annotations, corneal masks, eyelid/eyelash-line keypoints, and clinical tabular parameters for astigmatism-related research.

5,381 B-scan ocular ultrasound video clips with retinal-detachment presence and macula-on/off status labels. Total runtime approximately 5 hours 10 minutes. Only public ocular ultrasound video benchma

High-resolution multi-camera face images with calibrated gaze targets, extreme head poses, and varied illumination.

Fundus photography, multispectral and functional retinal imaging, retinal blood-flow and pupillary-light videos, retina-characteristic labels, quality labels, demographics, and psychological assessmen

Indoor and outdoor panoramic-camera recordings with continuous three-dimensional gaze labels across wide head poses and distances.

Crowdsourced iPhone and iPad face videos with screen-fixation coordinates for appearance-based mobile gaze estimation.

30,000 subjects (10K each AMD, DR, glaucoma) with paired SLO fundus and OCT B-scans, demographic attributes (race, ethnicity, gender, language), for fairness analysis.

45 high-resolution fundus images (healthy/DR/glaucoma) with manual vessel segmentation.

OCT challenge datasets from HDMILab covering retinal layer segmentation and fluid detection tasks from MICCAI/OMIA workshops. Includes sub-challenges such as AMD/CSC/DR classification and retinal laye

400 training fundus images with AMD classification (non-AMD vs AMD) and lesion annotations. Challenge dataset from ISBI 2020.

516 fundus images with DR grade (0-4), macular edema grade (0-2), and pixel-level lesion segmentation for 81 images.

Two COCO-format instance-segmentation datasets derived from cataract surgery videos, covering instruments and anatomical structures.

Multi-modal retinal vessel segmentation dataset spanning fluorescence angiography, fundus autofluorescence, and infrared imaging.

11,760 fundus images with glaucoma classification labels and ophthalmologist-derived attention maps. Largest public glaucoma dataset with attention ground truth.

Composite multimodal ophthalmology benchmark with multi-granular anatomical, diagnostic, staging, demographic, and text annotations.

High-speed head-mounted eye-region videos with pupil-center annotations under varied indoor, outdoor, eyewear, and lighting conditions.

1748 fundus images with DR grading (0-3 Retinopathy Grade) and macular edema risk (0-2).

34,452 fundus images from 4,344 patients across multiple fields per eye, with DR screening labels. Largest public four-field DR dataset.

Longitudinal laptop-camera images captured during natural daily use with gaze targets, face geometry, and normalized views.

Longitudinal near-infrared iris images with subject, eye, age, sex, and ethnicity metadata used in ICE iris-recognition evaluations.

A collection of 640 retinal OCT images with expert choroidal region annotations for choroid segmentation.

500 subjects with OCTA volumes, vessel segmentation, FAZ (foveal avascular zone) annotations, and layer segmentation. Largest public OCTA dataset.

500 OCT images: NORMAL (206), AMD (50), CSC (128), DR (59), MH (57). High-resolution B-scans for 5-class classification.

15 ultra-widefield (Optos) fundus images with pixel-level vessel segmentation ground truth.

148 Heidelberg Spectralis SD-OCT volumes (~4,254 B-scans) for 3-class volume-level classification: Normal (50 volumes), AMD (48 volumes), DME (50 volumes). Variable B-scans per volume: 19, 25, 31, or

One hundred twenty paired low quality and high quality clinical fundus photographs of the same eyes at 2560 by 2560 pixels for image restoration and enhancement research.

1200 fundus images: 400 train, 400 val, 400 test. Labels: glaucoma/non-glaucoma + optic disc/cup segmentation.

112 OCT volumes from three vendors (Cirrus / Spectralis / Topcon) with pixel-level segmentation of intraretinal fluid (IRF), subretinal fluid (SRF), and pigment epithelium detachment (PED).

40 fundus images (derived from DRIVE) with artery/vein annotations and vessel-tree structural labels.

100 fundus images with microaneurysm annotations from the Retinopathy Online Challenge.

Grand Challenge OCT classification dataset for diabetic-retinopathy related OCT classification.

400 macular OCT volumes; predict glaucoma Mean Deviation (MD, dB) from 24-2 Humphrey visual field test. Scalar regression task.

400 macular OCT volumes; predict 52-point Humphrey 24-2 visual field sensitivity map (0–100 dB per point). Multi-output regression.

400 macular OCT volumes; predict 52-point pattern deviation probability map from 24-2 Humphrey visual field test. Multi-output regression.

Restricted corneal confocal microscopy image set with pixel-level corneal-nerve masks annotated for high-precision segmentation.

Head-mounted eye images with pupil, iris, eyelid, eyeball, gaze, landmark, segmentation, and eye-movement annotations.

Visible-light iris images captured at a distance and on the move with realistic blur, reflection, occlusion, pose, and illumination noise.