A simple rotation classifier built with the mobilenet_small model to detected whether an ECG image was rotated or not and by how much. Only works for increments of 90 degrees. For more information and example of use see example.
The model is provided in the rotation_classifier folder while the training script and loops are in the training folder.
The model was trained on 1k real ECG images rotated artificially during training. The validation results for that dataset are as follows:
| Metric | Value |
|---|---|
| roc_auc | 0.999499 |
| precision | 0.987990 |
| recall | 0.988374 |
| f1 | 0.988025 |