(self, norm_nc, label_nc, config_text='spadeinstance3x3')
| 67 | # |label_nc|: the #channels of the input semantic map, hence the input dim of SPADE |
| 68 | class SPADE(nn.Module): |
| 69 | def __init__(self, norm_nc, label_nc, config_text='spadeinstance3x3'): |
| 70 | super().__init__() |
| 71 | |
| 72 | assert config_text.startswith('spade') |
| 73 | parsed = re.search('spade(\D+)(\d)x\d', config_text) |
| 74 | param_free_norm_type = str(parsed.group(1)) |
| 75 | ks = int(parsed.group(2)) |
| 76 | |
| 77 | self.param_free_norm = normalization(norm_nc) |
| 78 | |
| 79 | # The dimension of the intermediate embedding space. Yes, hardcoded. |
| 80 | nhidden = 128 |
| 81 | |
| 82 | pw = ks // 2 |
| 83 | self.mlp_shared = nn.Sequential( |
| 84 | nn.Conv2d(label_nc, nhidden, kernel_size=ks, padding=pw), |
| 85 | nn.ReLU() |
| 86 | ) |
| 87 | self.mlp_gamma = nn.Conv2d(nhidden, norm_nc, kernel_size=ks, padding=pw) |
| 88 | self.mlp_beta = nn.Conv2d(nhidden, norm_nc, kernel_size=ks, padding=pw) |
| 89 | |
| 90 | def forward(self, x_dic, segmap_dic, size=None): |
| 91 |
nothing calls this directly
no test coverage detected