Question

我收到此错误Error when checking input: expected conv2d_11_input to have 4 dimensions, but got array with shape (300, 300, 3) 如何将RGB图像传递到CNN？如何枚举样本以创建4D图像？

Answer 1

卷积函数需要一批图像，即使您有一个单一的图像，也应该将图像传递为：

#  assume img is your RGB image
#  add 4th batch dimension
img = np.expand_dims(img, axis=0)) 

#  now you can pass it to CNN

Answer 2

通常，要向numpy矩阵添加新维度，可以使用np.newaxis或np.expand_dims（或仅使用reshape）

import numpy as np 
nh, nw = 300, 300
x = np.random.rand(nh, nw, 3)

y = x[np.newaxis, ...]   
z = np.expand_dims(x, axis=0)

>>> x.shape
(300, 300, 3)
>>> y.shape
(1, 300, 300, 3)
>>> z.shape
(1, 300, 300, 3)

但是，如果您想将RGB馈入CNN模型，也许您还需要一步：交换频道并转换为float32：

import cv2 
import numpy as np 

fpath = "test.png"
nh, nw = 300, 300 

img = cv2.imread(fpath)
img = cv2.cvtColor(cv2.resize(img, (nw, nh)), cv2.COLOR_BGR2RGB)
ximg = np.float32(img) / 255.0 
ximg = ximg[np.newaxis, ...]

>>> (ximg.shape, ximg.dtype)
((1, 300, 300, 3), dtype('float32'))

如何在Python中将OpenCV图像输入经过训练的CNN模型（添加新维度）？

2 个答案: