README: 30 datasets
Browse files
README.md
CHANGED
|
@@ -96,6 +96,14 @@ configs:
|
|
| 96 |
data_files:
|
| 97 |
- split: train
|
| 98 |
path: sidecar/DRIVE/*.parquet
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 99 |
- config_name: FairSeg
|
| 100 |
data_files:
|
| 101 |
- split: train
|
|
@@ -352,7 +360,7 @@ configs:
|
|
| 352 |
|
| 353 |
# OphGen — curated public ophthalmic datasets
|
| 354 |
|
| 355 |
-
Curated copies of
|
| 356 |
images as stored in the OphGen image store (de-identified, de-duplicated, native resolution), the train / validation / test
|
| 357 |
split used in the project, six captions per image (short / medium / long × English / Chinese) and all original task annotations.
|
| 358 |
|
|
@@ -380,6 +388,7 @@ In `all`, an image released by several datasets appears once per dataset (same `
|
|
| 380 |
| [DDR](cards/DDR.md) | DDR | CFP | 6,552 | 2,688 | 4,232 | 3 |
|
| 381 |
| [DeepDRiD](cards/DeepDRiD.md) | DeepDRiD | CFP, UWF | 1,329 | 446 | 450 | 3 |
|
| 382 |
| [DRIVE](cards/DRIVE.md) | DRIVE | CFP | 18 | 2 | 20 | 1 |
|
|
|
|
| 383 |
| [FairSeg](cards/FairSeg.md) | Harvard-FairSeg | IR/NIR | 7,200 | 800 | 2,000 | 2 |
|
| 384 |
| [FIVES](cards/FIVES.md) | FIVES | CFP | 536 | 59 | 199 | 2 |
|
| 385 |
| [GAVE2](cards/GAVE2.md) | GAVE2 | FFA, CFP | 150 | 150 | 0 | 4 |
|
|
@@ -402,7 +411,7 @@ In `all`, an image released by several datasets appears once per dataset (same `
|
|
| 402 |
| [RVO-ME](cards/RVO-ME.md) | RVO-ME | OCT | 2,057 | 229 | 580 | 3 |
|
| 403 |
| [SJTU_Peripapillary](cards/SJTU_Peripapillary.md) | SJTU peripapillary OCT | OCT | 74 | 24 | 24 | 3 |
|
| 404 |
| [UWF-DR_2026](cards/UWF-DR_2026.md) | UWF-DR | UWF | 1,135 | 158 | 327 | 2 |
|
| 405 |
-
| **Total** | | | 363,
|
| 406 |
|
| 407 |
Split rules: `official` = official split; `official_val_from_train` = 10% of the official training set (by patient) used as
|
| 408 |
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
|
|
|
|
| 96 |
data_files:
|
| 97 |
- split: train
|
| 98 |
path: sidecar/DRIVE/*.parquet
|
| 99 |
+
- config_name: Duke_DME
|
| 100 |
+
data_files:
|
| 101 |
+
- split: train
|
| 102 |
+
path: data/Duke_DME/train-*.parquet
|
| 103 |
+
- split: validation
|
| 104 |
+
path: data/Duke_DME/validation-*.parquet
|
| 105 |
+
- split: test
|
| 106 |
+
path: data/Duke_DME/test-*.parquet
|
| 107 |
- config_name: FairSeg
|
| 108 |
data_files:
|
| 109 |
- split: train
|
|
|
|
| 360 |
|
| 361 |
# OphGen — curated public ophthalmic datasets
|
| 362 |
|
| 363 |
+
Curated copies of 30 public ophthalmic imaging datasets, prepared for the OphGen ophthalmic generative-model project:
|
| 364 |
images as stored in the OphGen image store (de-identified, de-duplicated, native resolution), the train / validation / test
|
| 365 |
split used in the project, six captions per image (short / medium / long × English / Chinese) and all original task annotations.
|
| 366 |
|
|
|
|
| 388 |
| [DDR](cards/DDR.md) | DDR | CFP | 6,552 | 2,688 | 4,232 | 3 |
|
| 389 |
| [DeepDRiD](cards/DeepDRiD.md) | DeepDRiD | CFP, UWF | 1,329 | 446 | 450 | 3 |
|
| 390 |
| [DRIVE](cards/DRIVE.md) | DRIVE | CFP | 18 | 2 | 20 | 1 |
|
| 391 |
+
| [Duke_DME](cards/Duke_DME.md) | Duke DME (Chiu) | OCT | 77 | 11 | 22 | 3 |
|
| 392 |
| [FairSeg](cards/FairSeg.md) | Harvard-FairSeg | IR/NIR | 7,200 | 800 | 2,000 | 2 |
|
| 393 |
| [FIVES](cards/FIVES.md) | FIVES | CFP | 536 | 59 | 199 | 2 |
|
| 394 |
| [GAVE2](cards/GAVE2.md) | GAVE2 | FFA, CFP | 150 | 150 | 0 | 4 |
|
|
|
|
| 411 |
| [RVO-ME](cards/RVO-ME.md) | RVO-ME | OCT | 2,057 | 229 | 580 | 3 |
|
| 412 |
| [SJTU_Peripapillary](cards/SJTU_Peripapillary.md) | SJTU peripapillary OCT | OCT | 74 | 24 | 24 | 3 |
|
| 413 |
| [UWF-DR_2026](cards/UWF-DR_2026.md) | UWF-DR | UWF | 1,135 | 158 | 327 | 2 |
|
| 414 |
+
| **Total** | | | 363,501 | 28,242 | 98,758 | |
|
| 415 |
|
| 416 |
Split rules: `official` = official split; `official_val_from_train` = 10% of the official training set (by patient) used as
|
| 417 |
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
|