def get_current_errors(self):
D_A_real = self.loss_D_A_real.item()
D_A_fake = self.loss_D_A_fake.item()
G_A = self.loss_G_A.item()
Cyc_A = self.loss_cycle_A.item()
D_B_real = self.loss_D_B_real.item()
D_B_fake = self.loss_D_B_fake.item()
G_B = self.loss_G_B.item()
Cyc_B = self.loss_cycle_B.item()
AE = self.loss_AE.item()
ret = OrderedDict([('D_A_real', D_A_real), ('D_A_fake', D_A_fake), ('G_A', G_A), ('Cyc_A', Cyc_A),
('D_B_real', D_B_real), ('D_B_fake', D_B_fake), ('G_B', G_B), ('Cyc_B', Cyc_B), ('AE', AE)])
if self.opt.identity > 0.0:
idt_A = self.loss_idt_A.item()
idt_B = self.loss_idt_B.item()
ret = OrderedDict(list(ret.items()) + [('idt_A', idt_A), ('idt_B', idt_B)])
if self.opt.lambda_gp > 0.0:
gp_A = self.loss_D_A_gp.item()
gp_B = self.loss_D_B_gp.item()
ret = OrderedDict(list(ret.items()) + [('D_A_gp', gp_A), ('D_B_gp', gp_B)])
if self.opt.lambda_color_mean > 0 or self.opt.lambda_color_sig_mean > 0:
ret = OrderedDict(list(ret.items()) + [('G_A_color', self.loss_color_A.data[0]), ('G_B_color', self.loss_color_B.data[0])])
if self.opt.log_grad:
g_D_A = util.get_grads(self.netD_A, ret_type='sum').item()
g_D_B = util.get_grads(self.netD_B, ret_type='sum').item()
g_E_A = util.get_grads(self.netE_A, ret_type='sum').item()
g_D = util.get_grads(self.net_D, ret_type='sum').item()
g_G_A = util.get_grads(self.netG_A, ret_type='sum').item()
g_E_C = util.get_grads(self.netE_C, ret_type='sum').item()
g_Dc = util.get_grads(self.net_Dc, ret_type='sum').item()
g_G_c = util.get_grads(self.netG_C, ret_type='sum').item()
g_G_B = util.get_grads(self.netG_B, ret_type='sum').item()
ret = OrderedDict(list(ret.items()) + [('D_A_grad', g_D_A), ('D_B_grad', g_D_B),('E_A_grad', g_E_A),('D_grad', g_D) ('G_A_grad', g_G_A), ('E_C_grad', g_E_C), ('G_C_grad', g_G_C), ('G_B_grad', g_G_B)])
return ret
def get_network_grads(self):
return [('E_A', util.get_grads(self.netE_A)),
('D', util.get_grads(self.net_D)),
('G_A', util.get_grads(self.netG_A)),
('E_C', util.get_grads(self.netE_C)),
('G_C', util.get_grads(self.netG_C)),
('G_B', util.get_grads(self.netG_B)),
('D_A', util.get_grads(self.netD_A)),
('D_B', util.get_grads(self.netD_B))]
def get_current_errors(self):
ret = OrderedDict([])
for i in range(len(self.loss_D_A_real)):
D_A_real = self.loss_D_A_real[i]
D_A_fake = self.loss_D_A_fake[i]
G_A = self.loss_G_A_list[i]
D_B_real = self.loss_D_B_real[i]
D_B_fake = self.loss_D_B_fake[i]
G_B = self.loss_G_B_list[i]
ret = OrderedDict(
list(ret.items()) +
[('D_A_real_%d' % i, D_A_real), ('D_A_fake_%d' % i, D_A_fake),
('G_A_%d' % i, G_A), ('D_B_real_%d' % i, D_B_real),
('D_B_fake_%d' % i, D_B_fake), ('G_B_%d' % i, G_B)])
if self.opt.lambda_gp > 0.0:
gp_A = self.loss_D_A_gp[i]
gp_B = self.loss_D_B_gp[i]
ret = OrderedDict(
list(ret.items()) +
[('D_A_gp_%d' % i, gp_A), ('D_B_gp_%d' % i, gp_B)])
for i in range(len(self.loss_cycle_A_list)):
Cyc_A = self.loss_cycle_A_list[i]
Cyc_B = self.loss_cycle_B_list[i]
ret = OrderedDict(
list(ret.items()) +
[('Cyc_A_%d' % i, Cyc_A), ('Cyc_B_%d' % i, Cyc_B)])
if self.opt.identity > 0.0:
idt_A = self.loss_idt_A.data[0]
idt_B = self.loss_idt_B.data[0]
ret = OrderedDict(
list(ret.items()) + [('idt_A', idt_A), ('idt_B', idt_B)])
if self.opt.alpha_gate != '' and i < len(self.loss_D_A_real) - 1:
if self.opt.alpha_gate == 'simp':
mask_A = self.netG_A.gate.mask
mask_B = self.netG_B.gate.mask
ret = OrderedDict(
list(ret.items()) +
[('G_A_mask', mask_A), ('G_B_mask', mask_B)])
elif 'trans' not in self.opt.alpha_gate:
mask_A = eval('self.netG_A.gate%d.mask.mean()' % i)
mask_B = eval('self.netG_B.gate%d.mask.mean()' % i)
ret = OrderedDict(
list(ret.items()) +
[('G_A_mask_%d' % i, mask_A.data[0]),
('G_B_mask_%d' % i, mask_B.data[0])])
if self.opt.lambda_color_mean > 0 or self.opt.lambda_color_sig_mean > 0:
ret = OrderedDict(
list(ret.items()) + [('G_A_color', self.loss_color_A.data[0]),
('G_B_color', self.loss_color_B.data[0])])
if self.opt.log_grad:
g_D_A = util.get_grads(self.netD_A, ret_type='sum').data[0]
g_D_B = util.get_grads(self.netD_B, ret_type='sum').data[0]
g_G_A = util.get_grads(self.netG_A, ret_type='sum').data[0]
g_G_B = util.get_grads(self.netG_B, ret_type='sum').data[0]
ret = OrderedDict(
list(ret.items()) + [('D_A_grad', g_D_A), ('D_B_grad', g_D_B),
('G_A_grad', g_G_A), ('G_B_grad', g_G_B)])
if self.opt.lambda_color_mean > 0 or self.opt.lambda_color_sig_mean > 0:
ret = OrderedDict(
list(ret.items()) + [('G_A_color', self.loss_color_A.data[0]),
('G_B_color', self.loss_color_B.data[0])])
if self.opt.lambda_style > 0:
ret = OrderedDict(
list(ret.items()) + [('G_A_style', self.style_loss_A.data[0]),
('G_B_style', self.style_loss_B.data[0])])
if self.opt.lambda_content > 0 or self.opt.lambda_content_l1 > 0:
ret = OrderedDict(
list(ret.items()) +
[('G_A_cont', self.content_loss_A.data[0]),
('G_B_cont', self.content_loss_B.data[0])])
return ret