摘抄一下MNIST手写体数据库文件格式

最近在查看Hinton最新的论文，提出了新的神经网络架构，最核心的神经元变成了一组取名为Capsule，输入相应的变成了向量（或者张量更有高大上的feel），关于该网络的实现已经有牛人复现了，感谢：

云梦居客（https://github.com/naturomics/CapsNet-Tensorflow）

https://www.zhihu.com/question/67287444/answer/251460831

在阅读代码时，发现了解析MNIST的文件，因此复习一下：

def load_mnist(path, is_training):    fd = open(os.path.join(cfg.dataset, 'train-images-idx3-ubyte'))    loaded = np.fromfile(file=fd, dtype=np.uint8)    trX = loaded[16:].reshape((60000, 28, 28, 1)).astype(np.float)    fd = open(os.path.join(cfg.dataset, 'train-labels-idx1-ubyte'))    loaded = np.fromfile(file=fd, dtype=np.uint8)    trY = loaded[8:].reshape((60000)).astype(np.int32)    fd = open(os.path.join(cfg.dataset, 't10k-images-idx3-ubyte'))    loaded = np.fromfile(file=fd, dtype=np.uint8)    teX = loaded[16:].reshape((10000, 28, 28, 1)).astype(np.float)    fd = open(os.path.join(cfg.dataset, 't10k-labels-idx1-ubyte'))    loaded = np.fromfile(file=fd, dtype=np.uint8)    teY = loaded[8:].reshape((10000)).astype(np.int32)    # normalization and convert to a tensor [60000, 28, 28, 1]    trX = tf.convert_to_tensor(trX / 255., tf.float32)    # => [num_samples, 10]    # trY = tf.one_hot(trY, depth=10, axis=1, dtype=tf.float32)    # teY = tf.one_hot(teY, depth=10, axis=1, dtype=tf.float32)    if is_training:        return trX, trY    else:        return teX / 255., teYdef get_batch_data():    trX, trY = load_mnist(cfg.dataset, cfg.is_training)    data_queues = tf.train.slice_input_producer([trX, trY])    X, Y = tf.train.shuffle_batch(data_queues, num_threads=cfg.num_threads,                                  batch_size=cfg.batch_size,                                  capacity=cfg.batch_size * 64,                                  min_after_dequeue=cfg.batch_size * 32,                                  allow_smaller_final_batch=False)    return(X, Y)

MNIST的idx格式，详细中文说明：http://www.jianshu.com/p/84f72791806f

MNIST数据库官网：http://yann.lecun.com/exdb/mnist/

FILE FORMATS FOR THE MNIST DATABASE

The data is stored in a very simple file format designed for storing vectors and multidimensional matrices. General info on this format is given at the end of this page, but you don't need to read that to use the data files.

All the integers in the files are stored in the MSB first (high endian) format used by most non-Intel processors. Users of Intel processors and other low-endian machines must flip the bytes of the header.

There are 4 files:

train-images-idx3-ubyte: training set images 
train-labels-idx1-ubyte: training set labels 
t10k-images-idx3-ubyte:  test set images 
t10k-labels-idx1-ubyte:  test set labels

The training set contains 60000 examples, and the test set 10000 examples.

The first 5000 examples of the test set are taken from the original NIST training set. The last 5000 are taken from the original NIST test set. The first 5000 are cleaner and easier than the last 5000.

TRAINING SET LABEL FILE (train-labels-idx1-ubyte):

[offset] [type]          [value]          [description] 
0000     32 bit integer  0x00000801(2049) magic number (MSB first) 
0004     32 bit integer  60000            number of items 
0008     unsigned byte   ??               label 
0009     unsigned byte   ??               label 
........ 
xxxx     unsigned byte   ??               label

The labels values are 0 to 9.

TRAINING SET IMAGE FILE (train-images-idx3-ubyte):

[offset] [type]          [value]          [description] 
0000     32 bit integer  0x00000803(2051) magic number 
0004     32 bit integer  60000            number of images 
0008     32 bit integer  28               number of rows 
0012     32 bit integer  28               number of columns 
0016     unsigned byte   ??               pixel 
0017     unsigned byte   ??               pixel 
........ 
xxxx     unsigned byte   ??               pixel

Pixels are organized row-wise. Pixel values are 0 to 255. 0 means background (white), 255 means foreground (black).

TEST SET LABEL FILE (t10k-labels-idx1-ubyte):

[offset] [type]          [value]          [description] 
0000     32 bit integer  0x00000801(2049) magic number (MSB first) 
0004     32 bit integer  10000            number of items 
0008     unsigned byte   ??               label 
0009     unsigned byte   ??               label 
........ 
xxxx     unsigned byte   ??               label

The labels values are 0 to 9.

TEST SET IMAGE FILE (t10k-images-idx3-ubyte):

[offset] [type]          [value]          [description] 
0000     32 bit integer  0x00000803(2051) magic number 
0004     32 bit integer  10000            number of images 
0008     32 bit integer  28               number of rows 
0012     32 bit integer  28               number of columns 
0016     unsigned byte   ??               pixel 
0017     unsigned byte   ??               pixel 
........ 
xxxx     unsigned byte   ??               pixel

Pixels are organized row-wise. Pixel values are 0 to 255. 0 means background (white), 255 means foreground (black). 
  

---

THE IDX FILE FORMAT

the IDX file format is a simple format for vectors and multidimensional matrices of various numerical types.

The basic format is

magic number 
size in dimension 0 
size in dimension 1 
size in dimension 2 
..... 
size in dimension N 
data

The magic number is an integer (MSB first). The first 2 bytes are always 0.

The third byte codes the type of the data: 
0x08: unsigned byte 
0x09: signed byte 
0x0B: short (2 bytes) 
0x0C: int (4 bytes) 
0x0D: float (4 bytes) 
0x0E: double (8 bytes)

The 4-th byte codes the number of dimensions of the vector/matrix: 1 for vectors, 2 for matrices....

The sizes in each dimension are 4-byte integers (MSB first, high endian, like in most non-Intel processors).

The data is stored like in a C array, i.e. the index in the last dimension changes the fastest. 
  
  

Happy hacking.