通过tf.data.dataset.from_generator迭代时出错

发布于 2025-02-09 11:44:52 字数 1680 浏览 3 评论 0原文

我想通过使用Python中的产量功能来构建自定义图像生产剂。 train_df 是具有两个列（ Input ：）的数据框，它包含通往我低分辨率图像的路径， upputs 列包含通往高分辨率图像的路径。 train_df dataframe的样本是在这里。将图像传递到批处理方式。

from numpy import asarray

def get_train_image(dataframe=train_df):

    for i in range(len(dataframe)):
        input_image = cv2.imread(dataframe.input[i],cv2.COLOR_BGR2RGB)
        input_image=cv2.resize(input_image,(80,120))
        input_image=asarray(input_image)
        input_image=input_image[np.newaxis,...]
        
        output_image = cv2.imread(dataframe.output[i],cv2.COLOR_BGR2RGB)
        output_image=cv2.resize(output_image,(120,160))
        output_image=asarray(output_image)
        output_image=output_image[np.newaxis,...]
        
        yield tf.convert_to_tensor(input_image),tf.convert_to_tensor(output_image)

作为图像生成器，当我使用下一个功能测试时。当我将数据适配器传递给CNN时，我会出错

tensorflow（tf.data.dataSet.from_generator（）以批量级的方式传递CNN的图像， .dataset.from_generator（get_train_image，output_signature =（tf.tensorspec（shape =（1,120,80,3），dtype = tf.float32），tf.tensorspec，tf.tensorspec（shape =（1,160,120,3），dtype =（1,160,120,3），args = [train_df]）

上面代码的图像：

错误的更多描述：

原文

I want to build a custom ImageGenerator by using yield function in Python. train_df is the dataframe which have two columns (input: that contains path to my low resolution image and output columns contains path to high resolution image. A sample of train_df dataframe is here. I made an yield function as Image Generator to pass the images into batchwise manner. Code for ImageGenerator:

from numpy import asarray

def get_train_image(dataframe=train_df):

    for i in range(len(dataframe)):
        input_image = cv2.imread(dataframe.input[i],cv2.COLOR_BGR2RGB)
        input_image=cv2.resize(input_image,(80,120))
        input_image=asarray(input_image)
        input_image=input_image[np.newaxis,...]
        
        output_image = cv2.imread(dataframe.output[i],cv2.COLOR_BGR2RGB)
        output_image=cv2.resize(output_image,(120,160))
        output_image=asarray(output_image)
        output_image=output_image[np.newaxis,...]
        
        yield tf.convert_to_tensor(input_image),tf.convert_to_tensor(output_image)

Image of above code Image_of_above_code. When I test this by using next function.It works fine Extracting_data_from_generator. I used DataAdapter provided by tensorflow (tf.data.Dataset.from_generator()) to pass the image in the CNN in batchwise manner and I got error while iterating. Same error is coming when I pass the Data Adaptor to CNN.

train_data=tf.data.Dataset.from_generator( get_train_image,output_signature=(tf.TensorSpec(shape=(1,120,80,3),dtype=tf.float32), tf.TensorSpec(shape=(1,160,120,3),dtype=tf.float32) ), args=[train_df])

Image of code above: