0
1

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 3 years have passed since last update.

Cognitive ServiceのCustom Visionで学習・テストを実行するPythonコード

Last updated at Posted at 2020-05-04

前提

  • AzureでCognitive ServiceのCustom Vision Serviceのリソースを作り、プロジェクトを作成しておく
  • プロジェクトの種類はClassificationにする
  • 学習用の画像を1タグあたり5枚以上用意し、タグごとにフォルダを作成して格納しておく
  • モデルをテストするための画像も別に用意しておく

学習用画像のアップロードと学習実行

train.py
import os, glob, time
from PIL import Image

from azure.cognitiveservices.vision.customvision.training import CustomVisionTrainingClient
from azure.cognitiveservices.vision.customvision.training.models import ImageFileCreateEntry

# 事前に以下の情報を環境変数に設定しておく
# export CUSTOMVISION_ENDPOINT=<your API endpoint>
# export CUSTOMVISION_TRAINING_KEY=<your training key>
# export CUSTOMVISION_PROJECT_ID=<your project id>

project_id = os.environ['CUSTOMVISION_PROJECT_ID']

trainer = CustomVisionTrainingClient(os.environ['CUSTOMVISION_TRAINING_KEY'], endpoint=os.environ['CUSTOMVISION_ENDPOINT'])

# タグのリストを作成
tag_list = []
tag_list.append(trainer.create_tag(project_id, "tagA"))
tag_list.append(trainer.create_tag(project_id, "tagB"))
tag_list.append(trainer.create_tag(project_id, "tagC"))

# タグに対応する画像のパスを指定
path_list = []
path_list.append('train-images/train/tagA/*.JPG')
path_list.append('train-images/train/tagB/*.JPG')
path_list.append('train-images/train/tagC/*.JPG')

# アップロード対象の画像リストを作成
image_list = []

for i, tag in enumerate(tag_list):
    for file_path in glob.glob(path_list[i]):
        with open(file_path, "rb") as image_contents:
            image_list.append(ImageFileCreateEntry(name=os.path.basename(file_path), contents=image_contents.read(), tag_ids=[tag.id]))

# 最大サイズの64件ずつアップロード
for i in range(0, len(image_list), 64):
    print('uploading', i, '-', i+64, '/', len(image_list))
    upload_result = trainer.create_images_from_files(project_id, images=image_list[i:i+64])

    if not upload_result.is_batch_successful:
        print("Image batch upload failed.")
        for image in upload_result.images:
            print("Image status: ", image.status)
        exit(-1)

# トレーニング実行
print ("Training...")
iteration = trainer.train_project(project_id)
while (iteration.status != "Completed"):
    iteration = trainer.get_iteration(project_id, iteration.id)
    print ("Training status: " + iteration.status)
    time.sleep(5)

作成したモデルをテスト用画像でテスト

quickTest.py
import os, glob
from PIL import Image

from azure.cognitiveservices.vision.customvision.training import CustomVisionTrainingClient
from azure.cognitiveservices.vision.customvision.training.models import ImageFileCreateEntry

def probability(prediction):
   return prediction.probability

# 事前に以下の情報を環境変数に設定しておく
# export CUSTOMVISION_ENDPOINT=<your API endpoint>
# export CUSTOMVISION_TRAINING_KEY=<your training key>
# export CUSTOMVISION_PROJECT_ID=<your project id>

project_id = os.environ['CUSTOMVISION_PROJECT_ID']

trainer = CustomVisionTrainingClient(os.environ['CUSTOMVISION_TRAINING_KEY'], endpoint=os.environ['CUSTOMVISION_ENDPOINT'])

# テストするIterationを指定
iterations = trainer.get_iterations(project_id)
target_iteration = iterations[0]

print('target iteration:', target_iteration.name)

# テストで使用するファイルを指定
files = glob.glob('train-images/test/**/*.JPG', recursive=True)

for i, file in enumerate(files):
    image = open(file, 'rb').read()
    result = trainer.quick_test_image(project_id, image, iteration_id=target_iteration.id)

    max_prediction = max(result.predictions, key=probability)

    print(file, max_prediction.tag_name, max_prediction.probability)

出力イメージ

target iteration: Iteration 1
train-images/test/tagA/IMG_001.JPG tagA 0.9999913
train-images/test/tagA/IMG_002.JPG tagA 0.9999975
train-images/test/tagB/IMG_003.JPG tagB 0.9999996
...

関連

pythonで機械学習用の画像を増幅する - Qiita

0
1
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
0
1

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?