Tensorflow flow_from_dataframe将所有像素设置为白色

Question

他，我在用flow_from_dataframe来管理图像，但是在计算结束的时候，我得到了所有错误的图像像素，下面我展示了我所有的代码，有人能帮我吗

batch_size = 12 
train_datagen = ImageDataGenerator(
rescale = 1. / 255,\
zoom_range=0.1,\
rotation_range=10,\
width_shift_range=0.1,\
height_shift_range=0.1,\
horizontal_flip=True,\
vertical_flip=False,)


train_generator_left_cc = train_datagen.flow_from_dataframe(
dataframe=df,
target_size=(256,256),
x_col="left_cc",
y_col=["lb","lm","rb","rm","left_cc"],
color_mode="rgb",
batch_size=batch_size,
class_mode="multi_output",
shuffle=False,
interpolation="bilinear"

)

train_generator_right_cc = train_datagen.flow_from_dataframe(
dataframe=df,
target_size=(256,256),
x_col="right_cc",
y_col=["lb","lm","rb","rm", "right_cc"],
color_mode="rgb",
batch_size=batch_size,
class_mode="multi_output",
shuffle=False,
)

x1,y1=train_generator_left_cc.next()
x2,y2=train_generator_right_cc.next()

i=11
image = x1[i]
label = (y1[0][i],y1[1][i],y1[2][i],y1[3][i],y1[4][i])  # categorical from one-hot-encoding
print("Image shape: ", image.shape)

print(label)
pyplot.imshow(image.squeeze())
pyplot.show()
image = x2[i]
label = (y2[0][i],y2[1][i],y2[2][i],y2[3][i]) # categorical from one-hot-encoding
print(label)
pyplot.imshow(image)
pyplot.show()

#i put the directory of the image in the label of the dataset to make test
image_decoded = tf.image.decode_png(tf.io.read_file(y2[4][i]), channels=3)


arr_ = np.squeeze(image_decoded) 
pyplot.imshow(arr_)
pyplot.show()

这是结果图像：result of the flow_from_dataframe

what i need

Answer 1

在张量流库中，有一个处理resize函数的bug，因此我为图像构建了一个自定义生成器