You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Comparative study of 7 CNN architectures for facial expression recognition using generic vs. Action Unit (FACS)-aligned facial-landmark regions — shared vs. landmark-specific backbones, feature-vector vs. feature-map fusion, and 3D convolution, evaluated on RAF-DB and CK+. IEEE CIS Kolkata internship project.
Controlled comparison of 7 models (ResNet50, EfficientNetB0, VGG16, MobileNetV2, custom CNN + KNN/SVM baselines) for 7-class facial expression recognition on CK+ under one identical training protocol.