Demo · EfficientNet-B7
See it classify test photos
These are the 81 test photos; none of them were used in training. Pick one to see what the model predicted, how confident it was about each class, and which parts of the photo drove its answer.
Preview data. The predictions and heatmaps on this page are placeholders while the models are being retrained. They will be replaced with EfficientNet-B7's real outputs.
True class Esotropia
CorrectPredicted: Esotropia
- Esotropia (true class)77%
- Exotropia4%
- Hypertropia3%
- Hypotropia10%
- Normal7%
Placeholder predictions — see the note at the top of the page.
How the glow is made. It is a Grad-CAM heatmap: it traces the model's answer back through its last layer of image features and highlights the regions that pushed it toward that class. A model that has learned the right thing should light up around the eyes.
Not a diagnosis. This is a research demo trained on 517 photos. It cannot tell you anything about your own eyes; an eye-care professional can.