# Data splits (chest X-ray)

- Fine-tuning split ID lists; download each dataset from its source under its own terms.
- `val` is for model selection.

| Dataset | Files | ID format | Train / Val / Test | Notes |
|---|---|---|---|---|
| ChestXray14 | `splits/chestxray14/` | image file name | 77,871 / 8,653 / 25,596 | Official test (patient-disjoint) |
| VinDr-CXR | `splits/vindr_cxr/` | VinDr image id | 13,494 / 1,506 / 3,000 | Official test |
| RSNA-Pneumonia | `splits/rsna_pneumonia/` | `patientId` | 21,346 / 2,669 / 2,669 | Patient-level split of the official train |
| SIIM-ACR | `splits/siim_acr/` | `SOPInstanceUID` | 9,303 / 1,372 / 1,372 | BenchX test split |
| ChestX-Det | `splits/chestx_det/` | image file name | 2,722 / 303 / 553 | Official test |
| MIMIC-CXR (report generation) | — | — | 27,079 / 2,130 / 3,858 | Official test, non-empty findings |