Pytorch转keras的有效方法,以FlowNet为例讲解

2023-07-24 05:50:03 152

Pytorch凭借动态图机制，获得了广泛的使用，大有超越tensorflow的趋势，不过在工程应用上，TF仍然占据优势。有的时候我们会遇到这种情况，需要把模型应用到工业中，运用到实际项目上，TF支持的PB文件和TF的C++接口就成为了有效的工具。今天就给大家讲解一下Pytorch转成Keras的方法，进而我们也可以获得Pb文件，因为Keras是支持tensorflow的，我将会在下一篇博客讲解获得Pb文件，并使用Pb文件的方法。

PytorchToKeras

首先，我们必须有清楚的认识，网上以及github上一些所谓的pytorch转换Keras或者Keras转换成Pytorch的工具代码几乎不能运行或者有使用的局限性（比如仅仅能转换某一些模型），但是我们是可以用这些转换代码中看出一些端倪来，比如二者的参数的尺寸（shape）的形式、channel的排序（firstorlast）是否一样，掌握到差异性，就能根据这些差异自己编写转换代码，没错，自己编写转换代码，是最稳妥的办法。整个过程也就分为两个部分。笔者将会以Nvidia开源的FlowNet为例，将开源的Pytorch代码转化为Keras模型。

按照Pytorch中模型的结构，编写对应的Keras代码，用keras的函数式API，构建起来会非常方便。

把Pytorch的模型参数，按照层的名称依次赋值给Keras的模型

以上两步虽然看上去简单，但实际我也走了不少弯路。这里一个关键的地方，就是参数的shape在两个框架中是否统一，那当然是不统一的。下面我以FlowNet为例。

Pytorch中的FlowNet代码

我们仅仅展示层名称和层参数，就不把整个结构贴出来了，否则会占很多的空间，形成水文。

先看用Keras搭建的flowNet模型，直接用model.summary()输出模型信息

__________________________________________________________________________________________________
Layer(type)OutputShapeParam#Connectedto
==================================================================================================
input_1(InputLayer)(None,6,512,512)0
__________________________________________________________________________________________________
conv0(Conv2D)(None,64,512,512)3520input_1[0][0]
__________________________________________________________________________________________________
leaky_re_lu_1(LeakyReLU)(None,64,512,512)0conv0[0][0]
__________________________________________________________________________________________________
zero_padding2d_1(ZeroPadding2D(None,64,514,514)0leaky_re_lu_1[0][0]
__________________________________________________________________________________________________
conv1(Conv2D)(None,64,256,256)36928zero_padding2d_1[0][0]
__________________________________________________________________________________________________
leaky_re_lu_2(LeakyReLU)(None,64,256,256)0conv1[0][0]
__________________________________________________________________________________________________
conv1_1(Conv2D)(None,128,256,25673856leaky_re_lu_2[0][0]
__________________________________________________________________________________________________
leaky_re_lu_3(LeakyReLU)(None,128,256,2560conv1_1[0][0]
__________________________________________________________________________________________________
zero_padding2d_2(ZeroPadding2D(None,128,258,2580leaky_re_lu_3[0][0]
__________________________________________________________________________________________________
conv2(Conv2D)(None,128,128,128147584zero_padding2d_2[0][0]
__________________________________________________________________________________________________
leaky_re_lu_4(LeakyReLU)(None,128,128,1280conv2[0][0]
__________________________________________________________________________________________________
conv2_1(Conv2D)(None,128,128,128147584leaky_re_lu_4[0][0]
__________________________________________________________________________________________________
leaky_re_lu_5(LeakyReLU)(None,128,128,1280conv2_1[0][0]
__________________________________________________________________________________________________
zero_padding2d_3(ZeroPadding2D(None,128,130,1300leaky_re_lu_5[0][0]
__________________________________________________________________________________________________
conv3(Conv2D)(None,256,64,64)295168zero_padding2d_3[0][0]
__________________________________________________________________________________________________
leaky_re_lu_6(LeakyReLU)(None,256,64,64)0conv3[0][0]
__________________________________________________________________________________________________
conv3_1(Conv2D)(None,256,64,64)590080leaky_re_lu_6[0][0]
__________________________________________________________________________________________________
leaky_re_lu_7(LeakyReLU)(None,256,64,64)0conv3_1[0][0]
__________________________________________________________________________________________________
zero_padding2d_4(ZeroPadding2D(None,256,66,66)0leaky_re_lu_7[0][0]
__________________________________________________________________________________________________
conv4(Conv2D)(None,512,32,32)1180160zero_padding2d_4[0][0]
__________________________________________________________________________________________________
leaky_re_lu_8(LeakyReLU)(None,512,32,32)0conv4[0][0]
__________________________________________________________________________________________________
conv4_1(Conv2D)(None,512,32,32)2359808leaky_re_lu_8[0][0]
__________________________________________________________________________________________________
leaky_re_lu_9(LeakyReLU)(None,512,32,32)0conv4_1[0][0]
__________________________________________________________________________________________________
zero_padding2d_5(ZeroPadding2D(None,512,34,34)0leaky_re_lu_9[0][0]
__________________________________________________________________________________________________
conv5(Conv2D)(None,512,16,16)2359808zero_padding2d_5[0][0]
__________________________________________________________________________________________________
leaky_re_lu_10(LeakyReLU)(None,512,16,16)0conv5[0][0]
__________________________________________________________________________________________________
conv5_1(Conv2D)(None,512,16,16)2359808leaky_re_lu_10[0][0]
__________________________________________________________________________________________________
leaky_re_lu_11(LeakyReLU)(None,512,16,16)0conv5_1[0][0]
__________________________________________________________________________________________________
zero_padding2d_6(ZeroPadding2D(None,512,18,18)0leaky_re_lu_11[0][0]
__________________________________________________________________________________________________
conv6(Conv2D)(None,1024,8,8)4719616zero_padding2d_6[0][0]
__________________________________________________________________________________________________
leaky_re_lu_12(LeakyReLU)(None,1024,8,8)0conv6[0][0]
__________________________________________________________________________________________________
conv6_1(Conv2D)(None,1024,8,8)9438208leaky_re_lu_12[0][0]
__________________________________________________________________________________________________
leaky_re_lu_13(LeakyReLU)(None,1024,8,8)0conv6_1[0][0]
__________________________________________________________________________________________________
deconv5(Conv2DTranspose)(None,512,16,16)8389120leaky_re_lu_13[0][0]
__________________________________________________________________________________________________
predict_flow6(Conv2D)(None,2,8,8)18434leaky_re_lu_13[0][0]
__________________________________________________________________________________________________
leaky_re_lu_14(LeakyReLU)(None,512,16,16)0deconv5[0][0]
__________________________________________________________________________________________________
upsampled_flow6_to_5(Conv2DTra(None,2,16,16)66predict_flow6[0][0]
__________________________________________________________________________________________________
concatenate_1(Concatenate)(None,1026,16,16)0leaky_re_lu_11[0][0]
leaky_re_lu_14[0][0]
upsampled_flow6_to_5[0][0]
__________________________________________________________________________________________________
inter_conv5(Conv2D)(None,512,16,16)4728320concatenate_1[0][0]
__________________________________________________________________________________________________
deconv4(Conv2DTranspose)(None,256,32,32)4202752concatenate_1[0][0]
__________________________________________________________________________________________________
predict_flow5(Conv2D)(None,2,16,16)9218inter_conv5[0][0]
__________________________________________________________________________________________________
leaky_re_lu_15(LeakyReLU)(None,256,32,32)0deconv4[0][0]
__________________________________________________________________________________________________
upsampled_flow5_to4(Conv2DTran(None,2,32,32)66predict_flow5[0][0]
__________________________________________________________________________________________________
concatenate_2(Concatenate)(None,770,32,32)0leaky_re_lu_9[0][0]
leaky_re_lu_15[0][0]
upsampled_flow5_to4[0][0]
__________________________________________________________________________________________________
inter_conv4(Conv2D)(None,256,32,32)1774336concatenate_2[0][0]
__________________________________________________________________________________________________
deconv3(Conv2DTranspose)(None,128,64,64)1577088concatenate_2[0][0]
__________________________________________________________________________________________________
predict_flow4(Conv2D)(None,2,32,32)4610inter_conv4[0][0]
__________________________________________________________________________________________________
leaky_re_lu_16(LeakyReLU)(None,128,64,64)0deconv3[0][0]
__________________________________________________________________________________________________
upsampled_flow4_to3(Conv2DTran(None,2,64,64)66predict_flow4[0][0]
__________________________________________________________________________________________________
concatenate_3(Concatenate)(None,386,64,64)0leaky_re_lu_7[0][0]
leaky_re_lu_16[0][0]
upsampled_flow4_to3[0][0]
__________________________________________________________________________________________________
inter_conv3(Conv2D)(None,128,64,64)444800concatenate_3[0][0]
__________________________________________________________________________________________________
deconv2(Conv2DTranspose)(None,64,128,128)395328concatenate_3[0][0]
__________________________________________________________________________________________________
predict_flow3(Conv2D)(None,2,64,64)2306inter_conv3[0][0]
__________________________________________________________________________________________________
leaky_re_lu_17(LeakyReLU)(None,64,128,128)0deconv2[0][0]
__________________________________________________________________________________________________
upsampled_flow3_to2(Conv2DTran(None,2,128,128)66predict_flow3[0][0]
__________________________________________________________________________________________________
concatenate_4(Concatenate)(None,194,128,1280leaky_re_lu_5[0][0]
leaky_re_lu_17[0][0]
upsampled_flow3_to2[0][0]
__________________________________________________________________________________________________
inter_conv2(Conv2D)(None,64,128,128)111808concatenate_4[0][0]
__________________________________________________________________________________________________
predict_flow2(Conv2D)(None,2,128,128)1154inter_conv2[0][0]
__________________________________________________________________________________________________
up_sampling2d_1(UpSampling2D)(None,2,512,512)0predict_flow2[0][0]

再看看Pytorch搭建的flownet模型

(conv0):Sequential(
(0):Conv2d(6,64,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(conv1):Sequential(
(0):Conv2d(64,64,kernel_size=(3,3),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(conv1_1):Sequential(
(0):Conv2d(64,128,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(conv2):Sequential(
(0):Conv2d(128,128,kernel_size=(3,3),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(conv2_1):Sequential(
(0):Conv2d(128,128,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(conv3):Sequential(
(0):Conv2d(128,256,kernel_size=(3,3),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(conv3_1):Sequential(
(0):Conv2d(256,256,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(conv4):Sequential(
(0):Conv2d(256,512,kernel_size=(3,3),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(conv4_1):Sequential(
(0):Conv2d(512,512,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(conv5):Sequential(
(0):Conv2d(512,512,kernel_size=(3,3),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(conv5_1):Sequential(
(0):Conv2d(512,512,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(conv6):Sequential(
(0):Conv2d(512,1024,kernel_size=(3,3),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(conv6_1):Sequential(
(0):Conv2d(1024,1024,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(deconv5):Sequential(
(0):ConvTranspose2d(1024,512,kernel_size=(4,4),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(deconv4):Sequential(
(0):ConvTranspose2d(1026,256,kernel_size=(4,4),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(deconv3):Sequential(
(0):ConvTranspose2d(770,128,kernel_size=(4,4),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(deconv2):Sequential(
(0):ConvTranspose2d(386,64,kernel_size=(4,4),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
(inter_conv5):Sequential(
(0):Conv2d(1026,512,kernel_size=(3,3),stride=(1,1),padding=(1,1))
)
(inter_conv4):Sequential(
(0):Conv2d(770,256,kernel_size=(3,3),stride=(1,1),padding=(1,1))
)
(inter_conv3):Sequential(
(0):Conv2d(386,128,kernel_size=(3,3),stride=(1,1),padding=(1,1))
)
(inter_conv2):Sequential(
(0):Conv2d(194,64,kernel_size=(3,3),stride=(1,1),padding=(1,1))
)
(predict_flow6):Conv2d(1024,2,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(predict_flow5):Conv2d(512,2,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(predict_flow4):Conv2d(256,2,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(predict_flow3):Conv2d(128,2,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(predict_flow2):Conv2d(64,2,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(upsampled_flow6_to_5):ConvTranspose2d(2,2,kernel_size=(4,4),stride=(2,2),padding=(1,1))
(upsampled_flow5_to_4):ConvTranspose2d(2,2,kernel_size=(4,4),stride=(2,2),padding=(1,1))
(upsampled_flow4_to_3):ConvTranspose2d(2,2,kernel_size=(4,4),stride=(2,2),padding=(1,1))
(upsampled_flow3_to_2):ConvTranspose2d(2,2,kernel_size=(4,4),stride=(2,2),padding=(1,1))
(upsample1):Upsample(scale_factor=4.0,mode=bilinear)
)
conv0Sequential(
(0):Conv2d(6,64,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv0.0Conv2d(6,64,kernel_size=(3,3),stride=(1,1),padding=(1,1))
conv0.1LeakyReLU(negative_slope=0.1,inplace)
conv1Sequential(
(0):Conv2d(64,64,kernel_size=(3,3),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv1.0Conv2d(64,64,kernel_size=(3,3),stride=(2,2),padding=(1,1))
conv1.1LeakyReLU(negative_slope=0.1,inplace)
conv1_1Sequential(
(0):Conv2d(64,128,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv1_1.0Conv2d(64,128,kernel_size=(3,3),stride=(1,1),padding=(1,1))
conv1_1.1LeakyReLU(negative_slope=0.1,inplace)
conv2Sequential(
(0):Conv2d(128,128,kernel_size=(3,3),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv2.0Conv2d(128,128,kernel_size=(3,3),stride=(2,2),padding=(1,1))
conv2.1LeakyReLU(negative_slope=0.1,inplace)
conv2_1Sequential(
(0):Conv2d(128,128,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv2_1.0Conv2d(128,128,kernel_size=(3,3),stride=(1,1),padding=(1,1))
conv2_1.1LeakyReLU(negative_slope=0.1,inplace)
conv3Sequential(
(0):Conv2d(128,256,kernel_size=(3,3),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv3.0Conv2d(128,256,kernel_size=(3,3),stride=(2,2),padding=(1,1))
conv3.1LeakyReLU(negative_slope=0.1,inplace)
conv3_1Sequential(
(0):Conv2d(256,256,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv3_1.0Conv2d(256,256,kernel_size=(3,3),stride=(1,1),padding=(1,1))
conv3_1.1LeakyReLU(negative_slope=0.1,inplace)
conv4Sequential(
(0):Conv2d(256,512,kernel_size=(3,3),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv4.0Conv2d(256,512,kernel_size=(3,3),stride=(2,2),padding=(1,1))
conv4.1LeakyReLU(negative_slope=0.1,inplace)
conv4_1Sequential(
(0):Conv2d(512,512,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv4_1.0Conv2d(512,512,kernel_size=(3,3),stride=(1,1),padding=(1,1))
conv4_1.1LeakyReLU(negative_slope=0.1,inplace)
conv5Sequential(
(0):Conv2d(512,512,kernel_size=(3,3),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv5.0Conv2d(512,512,kernel_size=(3,3),stride=(2,2),padding=(1,1))
conv5.1LeakyReLU(negative_slope=0.1,inplace)
conv5_1Sequential(
(0):Conv2d(512,512,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv5_1.0Conv2d(512,512,kernel_size=(3,3),stride=(1,1),padding=(1,1))
conv5_1.1LeakyReLU(negative_slope=0.1,inplace)
conv6Sequential(
(0):Conv2d(512,1024,kernel_size=(3,3),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv6.0Conv2d(512,1024,kernel_size=(3,3),stride=(2,2),padding=(1,1))
conv6.1LeakyReLU(negative_slope=0.1,inplace)
conv6_1Sequential(
(0):Conv2d(1024,1024,kernel_size=(3,3),stride=(1,1),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
conv6_1.0Conv2d(1024,1024,kernel_size=(3,3),stride=(1,1),padding=(1,1))
conv6_1.1LeakyReLU(negative_slope=0.1,inplace)
deconv5Sequential(
(0):ConvTranspose2d(1024,512,kernel_size=(4,4),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
deconv5.0ConvTranspose2d(1024,512,kernel_size=(4,4),stride=(2,2),padding=(1,1))
deconv5.1LeakyReLU(negative_slope=0.1,inplace)
deconv4Sequential(
(0):ConvTranspose2d(1026,256,kernel_size=(4,4),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
deconv4.0ConvTranspose2d(1026,256,kernel_size=(4,4),stride=(2,2),padding=(1,1))
deconv4.1LeakyReLU(negative_slope=0.1,inplace)
deconv3Sequential(
(0):ConvTranspose2d(770,128,kernel_size=(4,4),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
deconv3.0ConvTranspose2d(770,128,kernel_size=(4,4),stride=(2,2),padding=(1,1))
deconv3.1LeakyReLU(negative_slope=0.1,inplace)
deconv2Sequential(
(0):ConvTranspose2d(386,64,kernel_size=(4,4),stride=(2,2),padding=(1,1))
(1):LeakyReLU(negative_slope=0.1,inplace)
)
deconv2.0ConvTranspose2d(386,64,kernel_size=(4,4),stride=(2,2),padding=(1,1))
deconv2.1LeakyReLU(negative_slope=0.1,inplace)
inter_conv5Sequential(
(0):Conv2d(1026,512,kernel_size=(3,3),stride=(1,1),padding=(1,1))
)
inter_conv5.0Conv2d(1026,512,kernel_size=(3,3),stride=(1,1),padding=(1,1))
inter_conv4Sequential(
(0):Conv2d(770,256,kernel_size=(3,3),stride=(1,1),padding=(1,1))
)
inter_conv4.0Conv2d(770,256,kernel_size=(3,3),stride=(1,1),padding=(1,1))
inter_conv3Sequential(
(0):Conv2d(386,128,kernel_size=(3,3),stride=(1,1),padding=(1,1))
)
inter_conv3.0Conv2d(386,128,kernel_size=(3,3),stride=(1,1),padding=(1,1))
inter_conv2Sequential(
(0):Conv2d(194,64,kernel_size=(3,3),stride=(1,1),padding=(1,1))
)

因为Pytorch模型用name_modules()输出不是按顺序的，动态图机制决定了只有在有数据流动之后才知道走过的路径。所以上面的顺序也是乱的。但我想表明的是，我用Keras搭建的模型确实是根据官方开源的Pytorch模型搭建的。

模型搭建完毕之后，就到了关键的步骤：给Keras模型赋值。

给Keras模型赋值

这个步骤其实注意三个点

Pytorch是channels_first的，Keras默认是channels_last，在代码开头加上这两句：

K.set_image_data_format(‘channels_first')
K.set_learning_phase(0)

众所周知，卷积层的权重是一个4维张量，那么，在Pytorch和keras中，卷积核的权重的形式是否一致的，那自然是不一致的，要不然我为啥还要写这一点。那么就涉及到Pytorch权重的变形。

既然卷积层权重形式在两个框架是不一致的，转置卷积自然也是不一致的。

我们先看看卷积层在两个框架中的形式

keras的卷积层权重形式

我们用以下代码看keras卷积层权重形式

forlinmodel.layers:
print(l.name)
fori,winenumerate(l.get_weights()):
print('%d'%i,w.shape)

第一个卷积层输出如下0之后是卷积权重的shape，1之后的是偏置项

conv0
0(3,3,6,64)
1(64,)

所以Keras的卷积层权重形式是[height,width,input_channels,out_channels]

Pytorch的卷积层权重形式

net=FlowNet2SD()
forn,minnet.named_parameters():
print(n)
print(m.data.size())

conv0.0.weight
torch.Size([64,6,3,3])
conv0.0.bias
torch.Size([64])

用上面的代码得到所有层的参数的shape，同样找到第一个卷积层的参数，查看shape。

通过对比我们可以发现，Pytorch的卷积层shape是[out_channels,input_channels,height,width]的形式。

那么我们在取出Pytorch权重之后，需要用np.transpose改变一下权重的排序，才能送到Keras模型对应的层上。

Keras中转置卷积权重形式

deconv4
0(4,4,256,1026)
1(256,)

代码仍然和上面一样，找到转置卷积的对应的位置，查看一下

可以看出在Keras中，转置卷积形式是[height,width,out_channels,input_channels]

Pytorch中转置卷积权重形式

deconv4.0.weight
torch.Size([1026,256,4,4])
deconv4.0.bias
torch.Size([256])

代码仍然和上面一样，找到转置卷积的对应的位置，查看一下

可以看出在Pytorch中，转置卷积形式是[input_channels,out_channels,height,width]

小结

对于卷积层来说，Pytorch的权重需要使用

np.transpose(weight.data.numpy(),[2,3,1,0]）

才能赋值给keras模型对应的层的权重。

对于转置卷积来说，通过对比其实也是一样的。不信你去试试嘛。O(∩_∩)O哈哈~

对于偏置项，两种模块都是一维的向量，不需要处理。

有的情况还可能需要通道颠倒一下，但是很少需要这样做。

weights[::-1,::-1,:,:]

赋值

结束了预处理之后，我们就进入第二步，开始赋值了。

先看预处理的代码：

fork,vinweights_from_torch.items():
if'bias'notink:
weights_from_torch[k]=v.data.numpy().transpose(2,3,1,0)

赋值代码我只截了一部分供大家参考：

k_model=k_model()
forlayerink_model.layers:
current_layer_name=layer.name
ifcurrent_layer_name=='conv0':
weights=[weights_from_torch['conv0.0.weight'],weights_from_torch['conv0.0.bias']]
layer.set_weights(weights)
elifcurrent_layer_name=='conv1':
weights=[weights_from_torch['conv1.0.weight'],weights_from_torch['conv1.0.bias']]
layer.set_weights(weights)
elifcurrent_layer_name=='conv1_1':
weights=[weights_from_torch['conv1_1.0.weight'],weights_from_torch['conv1_1.0.bias']]
layer.set_weights(weights)

首先就是定义Keras模型，用layers获得所有层的迭代器。

遍历迭代器，对一个层赋予相应的值。

赋值需要用save_weights，其参数需要是一个列表，形式和get_weights的返回结果一致，即[conv_weights,bias_weights]

最后祝愿大家能实现自己模型的迁移。工程开源在了个人Github，有详细的使用介绍，并且包含使用数据，大家可以直接运行。

以上这篇Pytorch转keras的有效方法,以FlowNet为例讲解就是小编分享给大家的全部内容了，希望能给大家一个参考，也希望大家多多支持毛票票。

Pytorch转keras的有效方法,以FlowNet为例讲解

热门推荐

随机推荐