Diabetic Retinopathy Model Hub

Upload one retinal fundus image, run a selected model, or compare every registered DR model.

No image selected

DR Severity Mapping

0 No DR 1 Mild 2 Moderate 3 Severe 4 Proliferative DR
Results will appear here after inference.

Available Models

Entries are configuration-driven so structured HF repos, mixed repos, weights-only repos, and placeholders can live in one registry.

EfficientNetB5 AML

timm EfficientNet-B5 single-output regression model for diabetic retinopathy severity grading.

timm_safetensors / regression

Regression score decoded with tuned ordinal thresholds from the model repo.

ConvNeXt Tiny Ordinal

Ordinal ConvNeXt Tiny checkpoint with a structured HF subfolder and raw fallback weights.

mixed / ordinal

Ordinal outputs are decoded by thresholding cumulative DR-grade logits when applicable.

EfficientNetB4 + SwinBase Bilal

Parallel EfficientNetB4 + Swin Base feature extractor with a LayerNorm/Linear/GELU/Dropout/Linear fusion head trained with CORN ordinal loss.

weights_only / ordinal

Raw CORN ordinal checkpoint loaded with the original EfficientNetB4 + Swin Base parallel architecture.

DeiT With Distillation

Hugging Face-style DeiT retinal DR classifier with distillation.

structured_hf / classification

Uses a custom timm DeiT loader if the repo is not Transformers-native.

DeiT Without Distillation

Hugging Face-style DeiT retinal DR classifier without distillation.

structured_hf / classification

Does not use a custom timm DeiT loader

MobileNetV3 DR

Hugging Face-style MobileNetV3 diabetic retinopathy classifier.

structured_hf / classification

Uses a custom torchvision MobileNetV3 loader if Transformers auto-loading is not compatible.