Question

我正在一个项目中，在该项目中，我使用建立的keras模型对存储在Azure Blob存储中的图像进行分类，并将结果导出为.csv文件。通过使用get_blob_to_path并将一些图像下载到我的笔记本电脑上，我能够做到这一点。但是，由于图片太多，因此我希望不通过get_blob_to_bytes或get_blob_to_stream下载图片。

Answer 1

实际上，一种无需先下载就从Azure Blob存储加载图像的更简单的解决方案是使用sas令牌生成blob URL，以传递给imageio.imread。

这是我的代码从您的那里更改了。

from azure.storage.blob import BlockBlobService
from azure.storage.blob import ContainerPermissions
from datetime import datetime, timedelta

import imageio
import numpy as np
from skimage import transform
import pandas as pd

account_name = '<your account name>'
account_key = '<your account key>'
container_name = '<your container name>'

# generate the container-level sas token 

block_blob_service = BlockBlobService(account_name=account_name, account_key=account_key)
token = block_blob_service.generate_container_shared_access_signature(container_name, permission=ContainerPermissions.READ, expiry=datetime.utcnow() + timedelta(hours=1),)

# generate the list of blob urls with sas token

blob_names = service.list_blob_names(container_name)
df = pd.read_csv("~/Desktop/list.csv")
blob_urls_with_token = (f"https://{account_name}.blob.core.windows.net/{container_name}/{blob_name}?{token}" for blob_name in blob_names if blob_name in df.values)

#function to prepare the image for keras model

def load(img_sas_url):
    image = imageio.imread(img_sas_url)  # directly read image from the blob url with sas token
    image = np.array(image).astype('float32')/255
    image = transform.resize(image, (224, 224, 3))
    image = np.expand_dims(image, axis=0)
    return image

#predicting the images and append it to a datafram

predictions = []
images=[]
name = []
probs =[]
for img_sas_url in blob_urls_with_token:
    image = load(img_sas_url)
    prediction = model.predict_classes(image)
    prob = model.predict(image).max()
    predictions.append(prediction)
    probs.append(prob)
    images.append(file)
    name.append(root.split('\\')[4])

output = pd.DataFrame(
{'ImageID':name,
 'ImageName':images,
 'Predictions':predictions,
 'Probabilities':probs
})

希望有帮助。

使用Azure Blob存储中的图像作为输入

1 个答案: