云视觉api注释来自单个调用的多个映像。

Question

在我的图像处理项目中，我使用python、云视觉和云存储。在第一个链接中，google解释了单个图像处理，但我需要多个图像处理。我在云存储中有一个文件夹，我正在使用这些图像进行处理。

我尝试了第二和第三个环节，但它们是旧的答案，没有解决我的问题。我认为我应该为"input_image_uri“创建一个算法，比如将函数调用为for循环。我该怎么办，我不知道怎么走。

1- https://cloud.google.com/vision/docs/batch#vision_async_batch_annotate_images-python

2- How to annotate MULTIPLE images from a single call using Google's vision API? Python

3- https://www.py4u.net/discuss/247010

Answer 1

这是我使用的一种方法，它可以很好地从图像中进行OCR。我们将文件上传到一个桶文件夹中，获取该文件夹中所有图像的URIS，准备一个批处理请求并触发它。下面是一个例子。

import os

os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = "../my-creds.json" # If you are using service account
from google.cloud import storage
from google.cloud import vision


def file_list(bucket_name, bucket_dir: str):
    # list all files in bucket
    client: storage.Client = storage.Client()
    bucket = client.get_bucket(bucket_name)
    blob_list = list(bucket.list_blobs(prefix=bucket_dir, max_results=200))

    return [f"gs://{bucket_name}/{blob.name}" for blob in blob_list]


def prepare_request(file_uris: list):
    # Create the batch request
    requests = []
    # Example feature for OCR
    features = [vision.types.Feature(type=vision.enums.Feature.Type.DOCUMENT_TEXT_DETECTION)]

    for image in file_uris:
        source_image = vision.types.ImageSource(image_uri=image)
        request = vision.types.AnnotateImageRequest(image=vision.types.Image(source=source_image),
                                                    features=features)

        requests.append(request)
    return requests


def run_cloud_vision(gcv_credentials=None):
    client = vision.ImageAnnotatorClient(credentials=gcv_credentials)
    src_bucket = ""  # Your bucket name
    src_dir = "test-images/"  # relative path to your directory in bucket. Empty if no dir

    # get file URIs
    file_uris = file_list(src_bucket, src_dir)
    requests = prepare_request(file_uris)

    # trigger a batch annotation request
    response = client.batch_annotate_images(requests)

    # parse your response
    print("parsed response:", response)
    

if __name__ == '__main__':
    run_cloud_vision()

如果文件夹/桶中还有其他文件，这可能会列出您需要的其他文件。因此，您可能需要根据文件模式过滤URIS。

类似的方法也适用于其他视觉API。

Answer 2

我用局部方式用for循环来解决问题。

import os
from google.cloud import vision
import pandas as pd
import io

os.environ['GOOGLE_APPLICATION_CREDENTIALS'] = '.json key'

client = vision.ImageAnnotatorClient()
folder_path = 'main_folder_path'
df_main = pd.DataFrame(columns=['file_name', 'object', 'score'])

for data_file in sorted(os.listdir(folder_path)):
    path = folder_path + data_file
    # Image_source
    with io.open(path, 'rb') as image_file:
        content = image_file.read()

    image = vision.Image(content=content)

    response = client.object_localization(image=image)
    localized_object_annotations = response.localized_object_annotations

    df = pd.DataFrame(columns=['file_name', 'object', 'score'])
    for obj in localized_object_annotations:
        df = df.append(
            dict(
                file_name = data_file,
                object = obj.name,
                score = obj.score
            ), ignore_index=True
        )
    df_main = df_main.append(df)
df_main = df_main.to_excel('excel file path')