Strabismus

Step 5 · Classification

AlexNet

The baseline, used to understand how a network extracts features from this data.

The baseline, built to study feature extraction

From the paper

What we built

The AlexNet-based model was used to examine feature extraction and understand how the model operates. It incorporated DepthwiseConv2D layers, adaptive learning-rate scheduling and early stopping, mixed-precision training and the AdamW optimizer.

  • DepthwiseConv2D
  • Mixed precision
  • AdamW
  • Label smoothing
  • Adaptive learning rate
  • Early stopping
  1. Eye photo in
  2. Conv 11×11, 96 filtersstride 4
  3. Batch norm + max pool
  4. Depthwise conv 5×5
  5. Batch norm + max pool
  6. Conv 3×3 ×3384 · 384 · 256
  7. Max pool → global average pool
  8. Dense 4096 → dropout
  9. Dense 4096 → dropout
  10. Softmax, 5 classes
  11. EsotropiaExotropiaHypertropiaHypotropiaNormal

Published results

How AlexNet did on the 81 test photos

Accuracy
59.26%
Precision
58.88%
Recall
59.26%
F1 score
58.71%
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.
Predicted class →
True ↓EsotropiaExotropiaHypertropiaHypotropiaNormal
Esotropia0.730.070.070.000.13
Exotropia0.000.650.060.180.12
Hypertropia0.190.190.440.190.00
Hypotropia0.120.120.120.440.19
Normal0.060.060.060.120.71
01 — share of each true class; the diagonal is correct predictions

Each row is a true class; the numbers show where its test photos ended up. The diagonal is the share classified correctly.

It was strongest on esotropia (73% correct) and weakest on hypertropia (44% correct).