Results · published in the paper

How the four models compared

All numbers come from Table I and Figures 2–5 of our AIMLA 2025 paper. They were measured on the 81 test photos, which the models never saw during training.

EfficientNet-B784.00%
ResNet5074.00%
VGG1966.67%
AlexNet59.26%

Accuracy on the held-out test set, as published in the paper (Table I). Hover or focus a bar for all four metrics.

Every metric, every model
Model	Accuracy	Precision	Recall	F1 score
AlexNet	59.26	58.88	59.26	58.71
VGG19	66.67	67.45	66.67	67.00
ResNet50	74.00	79.00	73.00	74.00
EfficientNet-B7	84.00	85.00	83.00	84.40

Accuracy:: Out of all test photos, the share the model classified correctly.
Precision:: When the model names a class, how often it is right, averaged across classes.
Recall:: Of the photos that truly belong to a class, how many the model found, averaged across classes.
F1 score:: A single score that balances precision and recall; it is high only when both are.

Where each model went wrong

Confusion matrices

Each row is a true class and shows where its test photos ended up. A strong model puts its weight on the diagonal. The weaker models spread theirs: VGG19 and ResNet50, for example, send many of their mistakes to exotropia.

AlexNet

About this model →

Confusion matrix for AlexNet. Rows are the true class, columns the predicted class; each cell is the share of that true class's test photos.
True ↓	Eso	Exo	Hyper	Hypo	Normal
	Predicted class →
Eso	0.73	0.07	0.07	0.00	0.13
Exo	0.00	0.65	0.06	0.18	0.12
Hyper	0.19	0.19	0.44	0.19	0.00
Hypo	0.12	0.12	0.12	0.44	0.19
Normal	0.06	0.06	0.06	0.12	0.71

01 — share of each true class; the diagonal is correct predictions

VGG19

About this model →

Confusion matrix for VGG19. Rows are the true class, columns the predicted class; each cell is the share of that true class's test photos.
True ↓	Eso	Exo	Hyper	Hypo	Normal
	Predicted class →
Eso	0.80	0.13	0.00	0.00	0.07
Exo	0.06	0.47	0.18	0.24	0.06
Hyper	0.06	0.25	0.62	0.06	0.00
Hypo	0.06	0.12	0.06	0.69	0.06
Normal	0.00	0.18	0.06	0.00	0.76

01 — share of each true class; the diagonal is correct predictions

ResNet50

About this model →

Confusion matrix for ResNet50. Rows are the true class, columns the predicted class; each cell is the share of that true class's test photos.
True ↓	Eso	Exo	Hyper	Hypo	Normal
	Predicted class →
Eso	0.93	0.00	0.00	0.00	0.07
Exo	0.24	0.59	0.12	0.00	0.06
Hyper	0.00	0.06	0.88	0.06	0.00
Hypo	0.00	0.25	0.12	0.62	0.00
Normal	0.00	0.18	0.06	0.06	0.71

01 — share of each true class; the diagonal is correct predictions

EfficientNet-B7

About this model →

Confusion matrix for EfficientNet-B7. Rows are the true class, columns the predicted class; each cell is the share of that true class's test photos.
True ↓	Eso	Exo	Hyper	Hypo	Normal
	Predicted class →
Eso	0.80	0.07	0.00	0.07	0.07
Exo	0.06	0.76	0.00	0.06	0.12
Hyper	0.06	0.00	0.94	0.00	0.00
Hypo	0.00	0.06	0.06	0.75	0.12
Normal	0.00	0.00	0.00	0.06	0.94

01 — share of each true class; the diagonal is correct predictions

Next →See it classify test photos