(self, opt)
| 47 | return parser |
| 48 | |
| 49 | def __init__(self, opt): |
| 50 | BaseModel.__init__(self, opt) |
| 51 | |
| 52 | # specify the training losses you want to print out. |
| 53 | # The training/test scripts will call <BaseModel.get_current_losses> |
| 54 | self.loss_names = ['D_A', 'G_A', 'NCE1', 'D_B', 'G_B', 'NCE2', 'G'] |
| 55 | visual_names_A = ['real_A', 'fake_B'] |
| 56 | visual_names_B = ['real_B', 'fake_A'] |
| 57 | self.nce_layers = [int(i) for i in self.opt.nce_layers.split(',')] |
| 58 | |
| 59 | if opt.nce_idt and self.isTrain: |
| 60 | self.loss_names += ['idt_B', 'idt_A'] |
| 61 | visual_names_A.append('idt_B') |
| 62 | visual_names_B.append('idt_A') |
| 63 | |
| 64 | self.visual_names = visual_names_A + visual_names_B # combine visualizations for A and B |
| 65 | |
| 66 | if self.isTrain: |
| 67 | self.model_names = ['G_A', 'F1', 'D_A', 'G_B', 'F2', 'D_B'] |
| 68 | else: # during test time, only load G |
| 69 | self.model_names = ['G_A', 'G_B'] |
| 70 | |
| 71 | # define networks (both generator and discriminator) |
| 72 | self.netG_A = networks.define_G(opt.input_nc, opt.output_nc, opt.ngf, opt.netG, opt.normG, |
| 73 | not opt.no_dropout, opt.init_type, opt.init_gain, opt.no_antialias, |
| 74 | opt.no_antialias_up, self.gpu_ids, opt) |
| 75 | self.netG_B = networks.define_G(opt.input_nc, opt.output_nc, opt.ngf, opt.netG, opt.normG, |
| 76 | not opt.no_dropout, opt.init_type, opt.init_gain, opt.no_antialias, |
| 77 | opt.no_antialias_up, self.gpu_ids, opt) |
| 78 | self.netF1 = networks.define_F(opt.input_nc, opt.netF, opt.normG, |
| 79 | not opt.no_dropout, opt.init_type, opt.init_gain, opt.no_antialias, self.gpu_ids, |
| 80 | opt) |
| 81 | self.netF2 = networks.define_F(opt.input_nc, opt.netF, opt.normG, |
| 82 | not opt.no_dropout, opt.init_type, opt.init_gain, opt.no_antialias, self.gpu_ids, |
| 83 | opt) |
| 84 | |
| 85 | if self.isTrain: |
| 86 | self.netD_A = networks.define_D(opt.output_nc, opt.ndf, opt.netD, |
| 87 | opt.n_layers_D, opt.normD, opt.init_type, opt.init_gain, opt.no_antialias, |
| 88 | self.gpu_ids, opt) |
| 89 | self.netD_B = networks.define_D(opt.output_nc, opt.ndf, opt.netD, |
| 90 | opt.n_layers_D, opt.normD, opt.init_type, opt.init_gain, opt.no_antialias, |
| 91 | self.gpu_ids, opt) |
| 92 | self.fake_A_pool = ImagePool(opt.pool_size) # create image buffer to store previously generated images |
| 93 | self.fake_B_pool = ImagePool(opt.pool_size) # create image buffer to store previously generated images |
| 94 | # define loss functions |
| 95 | self.criterionGAN = networks.GANLoss(opt.gan_mode).to(self.device) |
| 96 | self.criterionNCE = [] |
| 97 | |
| 98 | for nce_layer in self.nce_layers: |
| 99 | self.criterionNCE.append(PatchNCELoss(opt).to(self.device)) |
| 100 | |
| 101 | self.criterionIdt = torch.nn.L1Loss().to(self.device) |
| 102 | self.criterionSim = torch.nn.L1Loss('sum').to(self.device) |
| 103 | self.optimizer_G = torch.optim.Adam(itertools.chain(self.netG_A.parameters(), self.netG_B.parameters()), |
| 104 | lr=opt.lr, betas=(opt.beta1, opt.beta2)) |
| 105 | self.optimizer_D = torch.optim.Adam(itertools.chain(self.netD_A.parameters(), self.netD_B.parameters()), |
| 106 | lr=opt.lr, betas=(opt.beta1, opt.beta2)) |
nothing calls this directly
no test coverage detected