Table 1 Description of each of the ten datasets considered in this paper in terms of image and population characteristics.
From: State-of-the-art retinal vessel segmentation with minimalistic models
Year | # ims. | Resolution | FOV | Challenges & Comments | |
|---|---|---|---|---|---|
STARE4 | 2000 | 20 | 605 \(\times \) 700 | 35\(^{\circ }\) | Poor quality: scanned and digitized photographs Healthy and pathological images (10/10) |
DRIVE1 | 2004 | 40 | 565 \(\times \) 584 | 45\(^{\circ }\) | Consistent good quality and contrast, low resolution Mostly healthy patients, some with mild DR (33/40) |
CHASE-DB 12 | 2012 | 28 | 999 \(\times \) 960 | 30\(^{\circ }\) | OD-centered images from 10-year old children Uneven background illumination and poor contrast |
HRF3 | 2013 | 45 | 3504 \(\times \) 2336 | 60\(^{\circ }\) | High visual quality, images taken with mydriatic dilation Healthy, diabetic, and glaucomatous patients (15/15/15) |
DRiDB9 | 2013 | 50 | 720 \(\times \) 576 | 45\(^{\circ }\) | Highly varying quality, illumination, and image noise Mostly diabetic patients of varying grades (36/50) |
AV-WIDE8 | 2015 | 30 | 2816 \(\times \) 1880 1500 \(\times \) 900 | 200\(^{\circ }\) | Uneven illumination, varying resolution due to cropping Healthy and age-related macular degeneration patients. |
IOSTAR6 | 2016 | 30 | 1024 \(\times \) 1024 | 45\(^{\circ }\) | Scanning Laser Ophthalmoscope images Macula-centered, high contrast and visual quality |
DR HAGIS7 | 2017 | 40 | 2816 \(\times \) 1880 4752 \(\times \) 3168 | 45\(^{\circ }\) | Multi-center, multi-device macula-centered images All diabetic patients with different co-morbities |
UoA-DR10 | 2017 | 200 | 2124 \(\times \) 2056 | 45\(^{\circ }\) | Both macula and OD-centered images Healthy, NP-DR and P-DR patients (56/114/30) |
LES-AV5 | 2018 | 22 | 1144 \(\times \) 1620 1958 \(\times \) 2196 | 30\(^{\circ }\) 45\(^{\circ }\) | OD-centered images, highly varying illumination 11 healthy and 11 glaucomatous patients |
