This project aims at evaluating the performances of the EfficientNet-b0 model while building an inference pipeline and analyzing the performances using different metrics.
For that task, we use a 50,000 images dataset. Images are equally distributed among 1000 classes.
Here are the global results (on the entire dataset and for all classes):
| Metrics | Values |
|---|---|
| Accuracy | 74.3% |
| Top 5 Accuracy | 91.9% |
| F1 Score | 74.3% |
| Precision | 74.3% |
| Recall | 74.3% |
| Specificity | 99.97% |
Please check the main.ipynb and classification_report.csv files for more results such as metrics' distributions per classes, analysis, interpretations and more.