因為程式改版, 改到忘記某一個正在訓練中的模型這個新增加的輸入值應該是什麼, 所以直接問 AI 問題,
下面模型的 use_res_skip 是 true or false?
UNetGenerator(
(embedder): Embedding(40, 128)
(model): UnetSkipConnectionBlock(
(submodule): UnetSkipConnectionBlock(
(submodule): UnetSkipConnectionBlock(
(submodule): UnetSkipConnectionBlock(
(submodule): UnetSkipConnectionBlock(
(att): SelfAttention(
(query_conv): Conv2d(512, 64, kernel_size=(1, 1), stride=(1, 1))
(key_conv): Conv2d(512, 64, kernel_size=(1, 1), stride=(1, 1))
(value_conv): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1))
)
(submodule): UnetSkipConnectionBlock(
(submodule): UnetSkipConnectionBlock(
(att): SelfAttention(
(query_conv): Conv2d(512, 64, kernel_size=(1, 1), stride=(1, 1))
(key_conv): Conv2d(512, 64, kernel_size=(1, 1), stride=(1, 1))
(value_conv): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1))
)
(down): Sequential(
(0): LeakyReLU(negative_slope=0.2, inplace=True)
(1): Conv2d(512, 512, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
)
(up): Sequential(
(0): ReLU(inplace=True)
(1): Conv2d(640, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(2): PixelShuffle(upscale_factor=2)
(3): InstanceNorm2d(512, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(down): Sequential(
(0): LeakyReLU(negative_slope=0.2, inplace=True)
(1): Conv2d(512, 512, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
(2): InstanceNorm2d(512, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
(up): Sequential(
(0): ReLU(inplace=True)
(1): Conv2d(512, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(2): PixelShuffle(upscale_factor=2)
(3): InstanceNorm2d(512, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(4): Dropout(p=0.5, inplace=False)
)
)
(down): Sequential(
(0): LeakyReLU(negative_slope=0.2, inplace=True)
(1): Conv2d(512, 512, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
(2): InstanceNorm2d(512, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
(up): Sequential(
(0): ReLU(inplace=True)
(1): Conv2d(512, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(2): PixelShuffle(upscale_factor=2)
(3): InstanceNorm2d(512, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(4): Dropout(p=0.5, inplace=False)
)
)
(down): Sequential(
(0): LeakyReLU(negative_slope=0.2, inplace=True)
(1): Conv2d(256, 512, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
(2): InstanceNorm2d(512, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
(up): Sequential(
(0): ReLU(inplace=True)
(1): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(2): PixelShuffle(upscale_factor=2)
(3): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(down): Sequential(
(0): LeakyReLU(negative_slope=0.2, inplace=True)
(1): Conv2d(128, 256, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
(2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
(up): Sequential(
(0): ReLU(inplace=True)
(1): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(2): PixelShuffle(upscale_factor=2)
(3): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(down): Sequential(
(0): LeakyReLU(negative_slope=0.2, inplace=True)
(1): Conv2d(64, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
(2): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
(up): Sequential(
(0): ReLU(inplace=True)
(1): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(2): PixelShuffle(upscale_factor=2)
(3): InstanceNorm2d(64, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(down): Sequential(
(0): Conv2d(1, 64, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
)
(up): Sequential(
(0): ReLU(inplace=True)
(1): Conv2d(64, 4, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(2): PixelShuffle(upscale_factor=2)
(3): Tanh()
)
)
)
[Network G] Total number of parameters : 52.875 M
Discriminator(
(model): Sequential(
(0): Conv2d(2, 64, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2))
(1): LeakyReLU(negative_slope=0.2, inplace=True)
(2): Conv2d(64, 128, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2))
(3): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(4): LeakyReLU(negative_slope=0.2, inplace=True)
(5): Conv2d(128, 256, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2))
(6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(7): LeakyReLU(negative_slope=0.2, inplace=True)
(8): Conv2d(256, 512, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(9): InstanceNorm2d(512, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(10): LeakyReLU(negative_slope=0.2, inplace=True)
(11): Conv2d(512, 1, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
)
(binary): Linear(in_features=2304, out_features=1, bias=True)
(catagory): Linear(in_features=2304, out_features=40, bias=True)
)
[Network D] Total number of parameters : 4.412 MAI 回:
根據你提供的模型結構輸出，
use_res_skip是 True。
判斷依據：
在 UNetGenerator 的 UnetSkipConnectionBlock 中，
use_res_skip決定了子模組（submodule）的輸出和編碼特徵（skip connection）是如何結合的：
- True (Addition): 直接相加
x + dec，通道數（Channels）維持與inner_nc相同。因此，上一層（該層的包覆層）的 up 卷積層輸入通道數會等於該層的inner_nc。 - False (Concatenation): 進行串接
cat([x, dec])，通道數變為inner_nc * 2。因此，上一層的 up 卷積層輸入通道數會是inner_nc * 2。
觀察你的模型輸出片段（以倒數第二層為例）：
(submodule): UnetSkipConnectionBlock(
...
(down): Sequential(
...
(1): Conv2d(256, 512, ...) <-- 此層 encoded (outer_nc=256, inner_nc=512)
)
(up): Sequential(
(0): ReLU(inplace=True)
(1): Conv2d(512, 1024, ...) <-- 關鍵在於這裡的輸入通道數是 512
...
)
)
- 此區塊的
inner_nc是 512（從down層的輸出可見）。 - 此區塊的 up 層輸入是 512。
- 如果
use_res_skip是False(Concatenation)，這裡的輸入應該是512 * 2 = 1024。 - 因為這裡輸入是 512（等於
inner_nc），表示下層回傳的特徵與 Skip Connection 是採相加方式，通道數未倍增。
所以
use_res_skip = True真棒, 提問問題並提供資料, 就可以無腦得到解答.
