该模型在 Tensorflow 到 CoreML 模型中没有有效的图像类型输入特征

Posted 2023-03-27

【中文标题】该模型在 Tensorflow 到 CoreML 模型中没有有效的图像类型输入特征

【英文标题】：The model does not have a valid input feature of type image in Tensorflow to CoreML model

【发布时间】：2021-11-19 06:19:49

【问题描述】：

我使用 Tensorflow 做了一个简单的图像分类 CNN。我想把它转换成 Core ML。我使用了 Core ML Tools 的通用转换 API。但是我的代码给了我以下错误：

模型没有图像类型的有效输入特征

这是 Xcode 中的模型详细信息。如何将模型的输入指定为图像而不是 MultiArray？

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.preprocessing import image
from tensorflow.keras.preprocessing.image import ImageDataGenerator
import numpy as np
import matplotlib.pyplot as plt

train_gen = ImageDataGenerator(rescale=1./255,shear_range=0.2,zoom_range=0.2,horizontal_flip=True)

test_gen = ImageDataGenerator(rescale=1./255,shear_range=0.2,zoom_range=0.2,horizontal_flip=True)

train_data = train_gen.flow_from_directory(directory='/Users/saibalaji/Documents/TensorFlowProjects/data/train',target_size=(256,256),batch_size=32)

train_data.class_indices

class_labels = []
for class_label,class_mode in train_data.class_indices.items():
    print(class_label)
    class_labels.append(class_label)


test_data = test_gen.flow_from_directory(directory='/Users/saibalaji/Documents/TensorFlowProjects/data/val',target_size=(256,256),batch_size=32)

model = tf.keras.Sequential()
model.add(tf.keras.layers.Conv2D(filters=32,kernel_size=(3,3),input_shape=(256,256,3),activation='relu'))
model.add(tf.keras.layers.MaxPool2D(pool_size=(2,2)))
model.add(tf.keras.layers.Conv2D(filters=64,kernel_size=(3,3),activation='relu'))
model.add(tf.keras.layers.MaxPool2D(pool_size=(2,2)))
model.add(tf.keras.layers.Conv2D(filters=128,kernel_size=(3,3),activation='relu'))
model.add(tf.keras.layers.MaxPool2D(pool_size=(2,2)))
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(units=256,activation='relu'))
model.add(tf.keras.layers.Dense(units=4,activation='softmax'))


model.summary()

model.compile(optimizer = 'adam',loss = 'categorical_crossentropy', metrics = ['accuracy'])

model.fit(x=train_data,validation_data=test_data,epochs=12)

def predict(filepath):
    test_image = image.load_img(path=filepath,target_size=(256,256))
    image_array = image.img_to_array(test_image)
    image_array = image_array / 255
    print(image_array.shape)
    image_array_exp = np.expand_dims(image_array,axis=0)
    result = model.predict(image_array_exp)
    print(class_labels[np.argmax(result)])
    plt.imshow(test_image)

predict('/Users/saibalaji/Documents/TensorFlowProjects/data/test/diseased cotton leaf/dis_leaf (322).jpg')

predict('/Users/saibalaji/Documents/TensorFlowProjects/data/test/fresh cotton leaf/d (341).jpg')


predict('/Users/saibalaji/Documents/TensorFlowProjects/data/test/diseased cotton plant/dd (367).jpg')

predict('/Users/saibalaji/Documents/TensorFlowProjects/data/test/diseased cotton plant/dd (706).jpg')

predict('/Users/saibalaji/Documents/TensorFlowProjects/data/test/fresh cotton plant/dsd (600).jpg')

model.save('cottoncnntwo.h5')

from tensorflow.keras.models import load_model
savedmodel = load_model('cottoncnn.h5')

savedmodel.summary()

import coremltools as ct

coremlmodel = ct.convert(savedmodel)


coremlmodel.save('cotton')

【参考方案1】：

请尝试：

image_input = ct.converters.mil.input_types.ImageType(name='input_image', shape=(256, 256, 3)) 
mlmodel_image = ct.convert(coremlmodel, inputs=[image_input])