keras list of Numpy arrays not the size model expe

问题:

I am having trouble finding the correct way of passing multiple inputs to a model. The model has 2 inputs

• noise image of shape (256, 256, 3)
• input image of shape (256, 256, 3)

and 1 output

• output image of shape (256, 256, 3)

I am producing the images via ImageDataGenerator:

x_data_gen = ImageDataGenerator(
    horizontal_flip=True,
    validation_split=0.2)

And I am producing the samples via a python generator:

def image_sampler(datagen, batch_size, subset="training"):

    for imgs in datagen.flow_from_directory('data/r_cropped', batch_size=batch_size, class_mode=None, seed=1, subset=subset):

        g_y = []

        noises = []
        bw_images = []
        for i in imgs:
            # append to expected output the original image
            g_y.append(i/255.0)

            noises.append(generate_noise(1, 256, 3)[0])
            bw_images.append(iu_rgb2gray(i))

        yield(np.array([noises, bw_images]), np.array(g_y))

When trying to train the model with:

    generator.fit_generator(
       image_sampler(x_data_gen, 32),
       validation_data=image_sampler(x_data_gen,32,"validation"),
       epochs=EPOCHS,
       steps_per_epoch= 540,
       validation_steps=160 )

I receive an error stating:

Error when checking model input: the list of Numpy arrays that you are passing to your model is not the size the model expected. Expected to see 2 array(s), but instead got the following list of 1 arrays

while the message is quite clear, I do not understand how to fix the generation process to solve it.

I tried:

    yield([noises, bw_images], np.array(g_y))

but this didn't work as it would reach a different error:

AttributeError: 'list' object has no attribute 'shape'

what am I missing?

回答1:

When you have multiple inputs/outputs you should pass them as a list of numpy arrays. So your second approach is correct but you have forgotten to convert the lists to numpy arrays in your second approach:

yield ([np.array(noises), np.array(bw_images)], np.array(g_y))

---

A more verbose approach to make sure everything is correct, is to choose names for the input and output layers. Example:

input_1 = layers.Input(# other args, name='input_1')
input_2 = layers.Input(# other args, name='input_2')

Then, use those names like this in your generator function:

yield ({'input_1': np.array(noises), 'input_2': np.array(bw_images)}, {'output': np.array(g_y)})

By doing so, you are making sure that the mapping is done correctly.