Tensorflow:Py_func返回未知形状 [英] Tensorflow: Py_func returns unknown shape

问题描述

我对tf.py_func函数有一个简单的问题.

I have a simple question re the tf.py_func function.

我有一个形状为(1,224,224,3)的图像张量my_img.为了测试py_func，我将张量输入到python函数return_tf中，该函数应返回相同的张量(按照文档转换为numpy数组后).

I have an image tensor my_img of shape (1,224,224,3). To test py_func, I feed the tensor to a python function return_tf that should give back the same tensor (after being converted to a numpy array as per docs).

代码如下:

def return_tf(x):
   return np.array(x)

test = tf.py_func(return_tf,[my_img],[tf.float32])

但是当我检查返回的张量的形状称为test时，我得到:

But when I checked the shape of the returned tensor called test, I get:

tf.Tensor 'PyFunc:0' shape=unknown dtype=float32

由于出现错误，我也无法在张量上运行eval():

I am also unable to run eval() on the tensor, since I get the error:

AttributeError: 'list' object has no attribute 'eval'.

任何人都知道如何修复tf.py_func返回的张量的张量形状吗?

Anyone knows how could I fix the tensor shape of the tensor returned by tf.py_func?

推荐答案

刚刚找到了解决方法..由于py_func返回张量流列表，我可以执行以下操作:

Just found a work-around.. since py_func returns a tensorflow list, I can do the ff:

test = tf.reshape(tf.concat(1, test), [ <<a shape>> ])

获得具有所需形状的张量

to get a tensor with a desired shape

这篇关于Tensorflow:Py_func返回未知形状的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！