As part of my bachelor's thesis on the classification of human emotions by facial image, I trained many models of the EfficientNetV2 family on FERPlus dataset.
Go to Releases to get trained models in ONNX format.
| Jupyter Notebook | Accuracy | Precision | Recall | F1 Score |
|---|---|---|---|---|
| 5_classes_EfficientNetV2B1_96x96_bs_256_weighted | 88.34% | 86.29% | 87.08% | 86.68% |
| 8_classes_EfficientNetV2B0_96x96_bs_256_weighted | 84.25% | 72.75% | 68.49% | 70.55% |
| 8_classes_EfficientNetV2B1_96x96_bs_256_weighted | 85.24% | 73.97% | 68.75% | 71.27% |
| 8_classes_EfficientNetV2S_96x96_bs_32 | 85.81% | 75.12% | 64.22% | 69.25% |
 |
 |
| 5_classes_EfficientNetV2B1_96x96_bs_256_weighted | 8_classes_EfficientNetV2B0_96x96_bs_256_weighted |
-
Download face-detection-retail-0044 model from Open Model Zoo repository and place it in the app/models folder.
-
Download a facial expression recognition model from Releases and place it in the app/models folder.
-
Run the program.
python app/app.py
