def forward(self, data, state):
x_interp, interp_score = data
gradient_x_interp = get_gradient(tf.reduce_sum(interp_score),
x_interp,
higher_order=True,
tape=state['tape'])
grad_l2 = tf.math.sqrt(
tf.reduce_sum(tf.math.square(gradient_x_interp), axis=[1, 2, 3]))
gp = tf.math.square(grad_l2 - 1.0)
return gp
def forward(self, data, state):
x_interp, interp_score = data
gradient_x_interp = get_gradient(torch.sum(interp_score), x_interp, higher_order=True)
grad_l2 = torch.sqrt(torch.sum(gradient_x_interp**2, dim=(1, 2, 3)))
gp = (grad_l2 - 1.0)**2
return gp