• A Riemannian metric learning regularization module for improving the model capacity is proposed.
• A hard sample reward strategy for refining the discriminability of the learning features is suggested.
If you want to train and evaluate the Grassmann network, run demo.py.
If you want to train and evaluate the Grassmann network with PM metric learning term, run demo_PM.py.
If you want to train and evaluate the Grassmann network with PAM metric learning term, run demo_PAM.py.