Two EfficientNet-B0 models trained for
student engagement detection.
-
daisee_imagenet_final_model.pth
Visual backbone trained on DAiSEE
Outputs 1280-dim feature vectors
Val F1: 0.4019 / Test F1: 0.3276
-
fer_for_daisee_transfer.pth
Facial expression concept extractor
Trained on FER2013
Used for NS ontology layer only
