如何在Python中读取MATLAB保存的字典数据？

我用MATLAB创建了字典，将字典的keys和values分别用-v7.3参数保存为.mat文件，MATLAB代码如下：

cnn6 = load("imgnet19k_less_feature_map_20230412_alexnet_cnn6_dict.mat");
dictionary_analyze = cnn6.dictionary_imagenames_features;
keys_in_order = keys(dictionary_analyze);
values_in_order = values(dictionary_analyze);
save("cnn6_keys_in_order_imagenet_19k_v2.mat", "keys_in_order", '-v7.3');
save("cnn6_values_in_order_imagenet_19k_v2.mat", "values_in_order", '-v7.3');

字典结构说明：

• 字典的key为字符串类型（对应图片名称）
• 每个value是二维数值数组（特征向量）

我尝试用Python读取这些.mat文件，代码如下：

import os
import sys
import pickle

import numpy as np
import h5py
import tables
from scipy.io import loadmat

normal_semantic_features_dir = "./normal_semantic_features/"

alexnet_features = ['cnn2', 'cnn4', 'cnn6', 'cnn8']

#dict_cnns = h5py.File(normal_semantic_features_dir + 'cnn2_keys_in_order_imagenet_19k_v2.mat', "r")
f1 = h5py.File(normal_semantic_features_dir + 'cnn8_keys_in_order_imagenet_19k_v2.mat', "r")
f2 = h5py.File(normal_semantic_features_dir + 'cnn8_values_in_order_imagenet_19k_v2.mat', "r")

array_of_keys = dict_cnns['keys_in_order']

print(list(f1.keys()))

运行结果：

> ['#refs#', '#subsystem#', 'keys_in_order']

    a = f1["keys_in_order"]
    type(a)

> h5py._hl.dataset.Dataset

    for k, v in annots.items():
        print(k,"........" ,annots[k])
        print("....")

> __header__ ........ b'MATLAB 5.0 MAT-file, Platform: PCWIN64, Created on: Fri Apr 14 13:32:47 2023' ....
> __version__ ........ 1.0 ....
> __globals__ ........ [] .... None ........ [(b'keys_in_order', b'MCOS', b'string', array([[3707764736],
>        [         2],
>        [         1],
>        [         1],
>        [         1],
>        [         1]], dtype=uint32))] ....
> __function_workspace__ ........ [[ 0  1 73 ...  0  0  0]]

请问如何读取这些.mat文件内容并转换为Python风格的字典？

MATLAB的-v7.3格式本质是HDF5格式，用h5py读取时，MATLAB的字符串数组、cell数组会以引用形式存储，需要特殊解析。以下是完整的读取和转换代码：

完整实现代码 #

import h5py
import numpy as np

def read_matlab_strings(h5_file, dataset_name):
    """解析MATLAB v7.3文件中的字符串数组"""
    ref_dataset = h5_file[dataset_name]
    # 获取所有字符串引用并扁平化
    str_refs = ref_dataset[()].flatten()
    keys = []
    for ref in str_refs:
        # 读取引用指向的字节数据，解码为UTF-8字符串
        str_bytes = h5_file[ref][()].tobytes()
        keys.append(str_bytes.decode('utf-8'))
    return keys

def read_matlab_values(h5_file, dataset_name):
    """解析MATLAB v7.3文件中的数值数组集合"""
    val_dataset = h5_file[dataset_name]
    val_refs = val_dataset[()].flatten()
    values = []
    for ref in val_refs:
        # 读取引用指向的数组，转置适配Python行优先存储
        arr = np.array(h5_file[ref[0]]).T
        values.append(arr)
    return values

# 文件路径配置
normal_semantic_features_dir = "./normal_semantic_features/"
key_file = f"{normal_semantic_features_dir}cnn8_keys_in_order_imagenet_19k_v2.mat"
val_file = f"{normal_semantic_features_dir}cnn8_values_in_order_imagenet_19k_v2.mat"

# 读取并转换为Python字典
with h5py.File(key_file, 'r') as f_keys, h5py.File(val_file, 'r') as f_vals:
    keys_list = read_matlab_strings(f_keys, 'keys_in_order')
    values_list = read_matlab_values(f_vals, 'values_in_order')
    python_dict = dict(zip(keys_list, values_list))

# 验证结果示例
print("前3个key：", keys_list[:3])
print("第一个value的形状：", python_dict[keys_list[0]].shape)

代码说明 #

1. 读取字符串keys：MATLAB的字符串在v7.3文件中以引用形式存储，需要遍历每个引用，读取对应的字节数据并解码为UTF-8字符串。
2. 读取数值values：字典的values是嵌套的数值数组，同样以引用形式存储，读取后转置数组（MATLAB是列优先存储，Python为行优先）。
3. 构建Python字典：用zip()将keys和values一一配对，直接转换为原生Python字典。

注意事项 #

• 确保依赖库已安装：执行pip install h5py numpy
• 如果values是其他类型的嵌套结构，需要根据MATLAB原数据结构调整read_matlab_values的解析逻辑。

内容的提问来源于stack exchange，提问作者Kadaj13