我想用tensorflow重写pytorch的torch.nn.functional.unfold功能:
#input x:[16, 1, 50, 36]
x = torch.nn.functional.unfold(x, kernel_size=(5, 36), stride=3)
#output x:[16, 180, 16]
tf.extract_image_patches() :x = tf.extract_image_patches(x,ksizes=[1, 1,5, 98],strides=[1, 1, 3, 1], rates=[1, 1, 1, 1],padding='VALID')输入 x.shape : [16,1,64,98]我得到输出 x.shape : [16,1,20,490]然后我 reshape X至[16,490,20] ，这是我所期望的。但是当我提供数据时出现错误:
UnimplementedError (see above for traceback): Only support ksizes across space.
[[Node:hcn/ExtractImagePatches = ExtractImagePatches[T=DT_FLOAT, ksizes=[1, 1, 5, 98], padding="VALID", rates=[1, 1, 1, 1], strides=[1, 1, 3, 1], _device="/job:localhost/replica:0/task:0/device:GPU:0"](hcn/Reshape)]]
torch.nn.functional.unfold更改X的功能?
最佳答案
x = tf.reshape(x, [16, 50, 36, 1])
x = tf.extract_image_patches(x, ksizes=[1, 4, 98, 1], strides=[1, 4, 1, 1], rates=[1, 1, 1, 1], padding='VALID')
关于python - 如何在 Tensorflow 中复制 PyTorch 的 nn.functional.unfold 函数？，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/64523441/