All shapes must be fully defined: [TensorShape([Dimension(299), Dimension(299), Dimension(3)])

All shapes must be fully defined: [TensorShape([Dimension(299), Dimension(299), Dimension(3)])

he data you want to batch must have pre-defined shape, in you case, tensor image doesn't, you need to specify the shape with image.set_shape or tf.image.resize_images




1. 在读入数据时,采用了如下代码形式

[cpp]  view plain  cop

1. def read_image(image_path):  
2.     """the image_path is the path of single image"""  
3.     file_contents = tf.read_file(input_queue[0])  
4.     image_data = tf.image.decode_jpeg(file_contents, channels=3)  
5.       
6.     label = input_queue[1]  
7.     image_id_wgb = input_queue[2]  
8.       
9.     return image_data, label, image_id_wgb  

[python]  view plain  copy

1. input_queue = tf.train.slice_input_producer([training_image_path_total,training_labels_total, training_image_id_total], shuffle=True)  
2. '''''the content of input_queue is the properties of one image, which includes a image_path(input_queue[0]), label(input_queue[1]), and image_id(input_queue[2]) of one image'''  
3.   
4. training_image_data, training_label, training_image_id= read_image(image_path = input_queue)  
5. training_image_batch, training_label_batch,training_image_id_batch = tf.train.shuffle_batch([training_image_data, training_label, training_image_id],batch_size=64,num_threads = 8, min_after_dequeue =101,capacity = 1000)
6. 
   
7. ValueError: All shapes must be fully defined: [TensorShape([Dimension(None), Dimension(None), Dimension(3)]), TensorShape([]), TensorShape([])]
8. 
   
9. 2. 解决办法</p>  
10. <在 tf.train.shuffle_batch  前加入了下面一个命令,才解决了.原因不知为何
11. training_image_data = tf.image.resize_images(training_image_data, [size, size])