def loss_func_rec_l1(x_out, t):
return F.mean_absolute_error(x_out, t)
python类mean_absolute_error()的实例源码
def pixel_wise_loss(self, x, y):
if self.loss_norm == 1:
return F.mean_absolute_error(x, y)
elif self.loss_norm == 2:
return F.mean_squared_error(x, y)
else:
raise ValueError('Invalid norm {}'.format(self.loss_norm))
def line_loss(self, x, t, k=3):
#print(x.data.type)
lx = x - F.max_pooling_2d(x, k, 1, 1)
lt = t - F.max_pooling_2d(t, k, 1, 1)
return 2 * F.mean_absolute_error(lx, lt)
# 0 for dataset
# 1 for fake
# G_p_rough: output of Generator (paired rough sketch)
# p_line: paired line art
def line_loss(self, x, t, k=3):
#print(x.data.type)
lx = x - F.max_pooling_2d(x, k, 1, 1)
lt = t - F.max_pooling_2d(t, k, 1, 1)
return 2 * F.mean_absolute_error(lx, lt)
# 0 for dataset
# 1 for fake
# G_p_rough: output of Generator (paired rough sketch)
# p_line: paired line art
def loss_G(self, real_B, fake_B, fake_D):
loss_l1 = F.mean_absolute_error(real_B, fake_B)
chainer.report({'loss_l1': loss_l1}, self.G)
batch_size, _, h, w = fake_D.shape
loss_D = - F.sum(F.log(fake_D + self.eps)) / (batch_size*h*w)
chainer.report({'loss_D': loss_D}, self.G)
loss = loss_D + self.lambd*loss_l1
chainer.report({'loss': loss}, self.G)
return loss
def loss_enc(self, enc, x_out, t_out, y_out, lam1=100, lam2=1):
batchsize,_,w,h = y_out.data.shape
loss_rec = lam1*(F.mean_absolute_error(x_out, t_out))
loss_adv = lam2*F.sum(F.softplus(-y_out)) / batchsize / w / h
loss = loss_rec + loss_adv
chainer.report({'loss': loss}, enc)
return loss
def loss_dec(self, dec, x_out, t_out, y_out, lam1=100, lam2=1):
batchsize,_,w,h = y_out.data.shape
loss_rec = lam1*(F.mean_absolute_error(x_out, t_out))
loss_adv = lam2*F.sum(F.softplus(-y_out)) / batchsize / w / h
loss = loss_rec + loss_adv
chainer.report({'loss': loss}, dec)
return loss
def __call__(self, x):
h = x
h = F.leaky_relu(self.c0(h))
h = F.leaky_relu(self.c1(h))
h = F.leaky_relu(self.c2(h))
h = F.leaky_relu(self.c3(h))
h = F.leaky_relu(self.l4(h))
h = F.reshape(F.leaky_relu(self.l5(h)),
(x.data.shape[0], self.ch, 4, 4))
h = F.leaky_relu(self.dc3(h))
h = F.leaky_relu(self.dc2(h))
h = F.leaky_relu(self.dc1(h))
h = F.tanh(self.dc0(h))
return F.mean_absolute_error(h, x)
def __init__(self, model, lossfunc=F.mean_absolute_error ):
super(LossEval, self).__init__()
self.lossfunc = lossfunc
with self.init_scope():
self.model = model
def loss_cnn(self, cnn, x_out, dst, dis_out, lam1=100, lam2=1):
loss_rec = lam1 * ( F.mean_absolute_error(x_out, dst) )
batchsize,_,w,h = dis_out.data.shape
loss_adv = lam2 * F.sum( F.softplus(-dis_out) ) / batchsize / w / h
loss = loss_rec + loss_adv
chainer.report({'loss': loss,"loss_rec":loss_rec, 'loss_adv': loss_adv }, cnn)
return loss
