Skip to content

Latest commit

 

History

History
118 lines (84 loc) · 5.06 KB

CollectDataForRe-training.md

File metadata and controls

118 lines (84 loc) · 5.06 KB
Raw

Collect data for model retraining

Batch-upload new training data

$ pip3 install azure-cognitiveservices-vision-customvision
$ pip3 install msrest

You can find other Cognitive Services modules here.

  • Secondly, if you label your images in PASCAL VOC format and save them as .xml files, you have to convert the original format to the format of Azure Custom Vision. Here lists the sample codes to convert 4 labels (people, foam, hand, hat).
def process_data_for_azure(filepath, dir_name):
    xml_files = glob.glob(os.path.join(filepath, '*.xml'))
    image_files = glob.glob(os.path.join(filepath, '*.jpg'))
    dir_image = {}
    for xml, img in zip(xml_files, image_files):
        tree = ET.parse(xml)
        root = tree.getroot()

        w = root.find("size")[0].text
        h = root.find("size")[1].text

        img_name = img.split('\\')[-1][:-4]

        people_region = []
        foam_region = []
        hand_region = []
        hat_region = []

        temp = {}

        for item in root.findall("object"):
            class_name = item[0].text.replace('-', '')

            x = int(item[4][0].text) / float(w)
            y = int(item[4][1].text) / float(h)
            dw = (int(item[4][2].text) - int(item[4][0].text)) / float(w)
            dh = (int(item[4][3].text) - int(item[4][1].text)) / float(h)

            if class_name == "people":
                people_region.append([x, y, dw, dh])
            if class_name == "foam":
                foam_region.append([x, y, dw, dh])
            if class_name == "hand":
                hand_region.append([x, y, dw, dh])
            if class_name == "hat":
                hat_region.append([x, y, dw, dh])

        temp["people"] = People_region
        temp["foam"] = Foam_region
        temp["hand"] = hand_1_region
        temp["hat"] = hat_region
        
        dir_image[img_name] = temp

    images[dir_name] = dir_image
    return len(dir_image)
  • Finally, prepare your data, and there are four cases for uploading your data. These cases arw shown below.
new project existed project
labeled images case 1 case 2
unlabeled images case 3 case 4

NOTICE: An existed project without any labeled images or any tags is not allowed in this practice.

NOTICE: tags in azure_cv_config.ini must be consistant with tags in Azure Custom Vision.

case 1. Uploading labelled data to a new project

  • In this case, you don't have to add a new project manually. You just fill out all items except prediction_key and project_id in the azure_cv_config.ini, and your project will be built.

  • training_key and ENDPOINT can be found in your custom vision in Azure Protal.

case 2. Uploading labelled data to an existed project

  • In this case, fill The azure_cv_config.ini, out and then execute the python codes.

  • Required information can be found in the Settings, where is located by red arrow, of your Azure Custom Vision project.

case 3. Uploading unlabelled data to a new project

case 4. Uploading unlabelled data to an existed project

Take Uploading labelled data to a new project for example, here is the command to execute upload_labelled_data_in_new_project.py.

$ python3 upload_labelled_data_in_new_project.py