RetinAI — ROP model weights
Trained checkpoints for the RetinAI BE capstone project (VESIT, 2026–27).
Downloaded automatically by scripts/get_weights.py in the
project repository.
| File | Model | Notes |
|---|---|---|
rop_screening_resnet50.pth |
ROP binary screening (ResNet50) | AUC 0.927 pooled / 0.881 device-controlled; sens 0.990 @ thr 0.1933 |
rop_staging_effnetv2s_structured_fold0.pth |
6-class ICROP staging, research preview | Structured head (ordinal CORN + AP-ROP branch); fold-0 checkpoint of a 5-fold CV (macro-F1 0.692 ± 0.086) |
Not a medical device. Research prototype, no regulatory clearance, no prospective validation. The staging checkpoint is a research preview, not a clinical grade. Trained on de-identified images; this repository contains weights only, no patient data.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support