Python LMDB

读取LMDB文件

# 打开LMDB数据库
env = lmdb.open('/usr/local/project/dataset/lmdb', readonly=True)

txn = env.begin()

count = 1
# 通过cursor()遍历所有数据和键值
for key, value in txn.cursor():
    k = key.decode('utf-8')
    if 'image' in k:
        continue
    print(key)
    print(value)
    print(value.decode('utf-8'))
    # 使用 `io.BytesIO` 将图片流转换为文件对象
    # image_bytes = io.BytesIO(value)
    # 使用 Pillow 打开图片
    # image = Image.open(image_bytes)
    # image.show()
    count += 1
    if count <= 3:
        continue
    break


env.close()

创建 LMDB文件

import lmdb

# 创建LMDB环境
env = lmdb.open('mydatabase.lmdb', map_size=1099511627776)  # 1TB map size

# 获取事务对象
with env.begin(write=True) as txn:
    # 添加数据
    txn.put(key='my_key'.encode(), value=b'my_value')

# 关闭环境
env.close()

获取图片流

def get_image_io(url):
    with open(url,'rb') as f:
        return f.read()

创建lmdb数据集 CVS -> lmdb

import lmdb
import pandas as pd

# 读取CSV文件
data = pd.read_csv('ICDAR13_HCTR_Dataset.csv')

# 打印数据
imagePathArr = data.ImagePath
labelArr = data.Label


def get_image_io(url):
    with open(url,'rb') as f:
        return f.read()

# # 创建LMDB环境
env = lmdb.open('mydatabase.lmdb', map_size=314572800)
i = 0
# 获取事务对象
with env.begin(write=True) as txn:
    # 添加数据
    while i < len(data.ImagePath):
        i += 1
        indexStr = str(i).zfill(9)
        imgKey = f'image-{indexStr}'.encode(encoding='UTF-8', errors='strict')
        labelKey = f'label-{indexStr}'.encode(encoding='UTF-8', errors='strict')
        imgIo = get_image_io(imagePathArr[i - 1])
        text = labelArr[i - 1]
        text = text.encode(encoding='UTF-8', errors='strict')
        txn.put(key=labelKey, value=text)
        txn.put(key=imgKey, value=imgIo)
    txn.put(key="num-samples".encode(encoding='UTF-8', errors='strict'),value=str(len(imagePathArr)).encode(encoding='UTF-8', errors='strict'))
env.close()