可以在model.predict()之前将preprocess_input函数应用于测试集吗?我在训练和验证集上使用了preprocess_input函数,并且如果我不对测试集应用preprocess_input,该模型会预测效果很差。而当我将其应用于预测之前的测试集时,它的预测非常准确。 任何建议将不胜感激!
train_datagen = ImageDataGenerator(preprocessing_function=tf.keras.applications.densenet.preprocess_input)
train_generator = train_datagen.flow_from_directory(train_dir,
batch_size=16,
class_mode='categorical',
target_size=(224, 224)
)
validation_datagen=ImageDataGenerator(preprocessing_function=tf.keras.applications.densenet.preprocess_input)
validation_generator = validation_datagen.flow_from_directory(val_dir,
batch_size=8,
class_mode='categorical',
target_size=(224, 224))
testGenerator=ImageDataGenerator(preprocessing_function=tf.keras.applications.densenet.preprocess_input).flow_from_directory(test_dir,batch_size=16,class_mode='categorical',shuffle=False,target_size=(224,224))