(self, in_channels, out_channels, ch, num_res_blocks, resolution,
ch_mult=(2, 2), dropout=0.0)
| 1276 | |
| 1277 | class UpsampleDecoder(nn.Module): |
| 1278 | def __init__(self, in_channels, out_channels, ch, num_res_blocks, resolution, |
| 1279 | ch_mult=(2, 2), dropout=0.0): |
| 1280 | super().__init__() |
| 1281 | # upsampling |
| 1282 | self.temb_ch = 0 |
| 1283 | self.num_resolutions = len(ch_mult) |
| 1284 | self.num_res_blocks = num_res_blocks |
| 1285 | block_in = in_channels |
| 1286 | curr_res = resolution // 2 ** (self.num_resolutions - 1) |
| 1287 | self.res_blocks = nn.ModuleList() |
| 1288 | self.upsample_blocks = nn.ModuleList() |
| 1289 | for i_level in range(self.num_resolutions): |
| 1290 | res_block = [] |
| 1291 | block_out = ch * ch_mult[i_level] |
| 1292 | for i_block in range(self.num_res_blocks + 1): |
| 1293 | res_block.append(ResnetBlock(in_channels=block_in, |
| 1294 | out_channels=block_out, |
| 1295 | temb_channels=self.temb_ch, |
| 1296 | dropout=dropout)) |
| 1297 | block_in = block_out |
| 1298 | self.res_blocks.append(nn.ModuleList(res_block)) |
| 1299 | if i_level != self.num_resolutions - 1: |
| 1300 | self.upsample_blocks.append(Upsample(block_in, True)) |
| 1301 | curr_res = curr_res * 2 |
| 1302 | |
| 1303 | # end |
| 1304 | self.norm_out = Normalize(block_in) |
| 1305 | self.conv_out = torch.nn.Conv2d(block_in, |
| 1306 | out_channels, |
| 1307 | kernel_size=3, |
| 1308 | stride=1, |
| 1309 | padding=1) |
| 1310 | |
| 1311 | def forward(self, x): |
| 1312 | # upsampling |
nothing calls this directly
no test coverage detected