Importing Brush Traces

You may import PNG masks as pre-labels for the Brush labeling tool in Ango Hub.

How to Import Masks as Brush Annotations

1. Verify Brush Tool Classes

Navigate to Settings > Category Schema in your project to ensure that the necessary Brush tool classes are available in the project.

2. Assign Colors

• For instance segmentation tasks, randomly assign a unique RGB color value to each instance in your mask.

• For semantic segmentation tasks, randomly assign a unique RGB color value to each class in your mask.

3. Prepare Masks

Prepare the masks you intend to import. Ensure the masks meet the following criteria:

• The imported mask must be a 4-channel RGBA image

• The width and height of the mask image must match the image asset's dimensions

• The background value of the mask must be solid white and transparent

  • Example: (RGBA: 255, 255, 255, 0)

• Use randomly assigned unique colors in the previous step, with fully opaque (255) alpha channel. Make sure that the mask’s foreground brush content must match the exact color values of the "brush" field in the JSON file.

  • Example:

    • Mask Image: (RGBA: 244, 67, 54, 255),

    • JSON File: "brush": [244, 67, 54]

The following would be a valid PNG mask:

6KB

brush-import-docs.png

image

You can use the following Python script to convert an RGB mask with a solid black background into the required RGBA format:

import numpy as np
from skimage import io

def rgb_mask_to_rgba_mask(input_mask_path, output_mask_path):
    input_mask = io.imread(input_mask_path)
    output_mask = np.zeros((input_mask.shape[0], input_mask.shape[1], 4), dtype=np.uint8)

    mask_flat = input_mask.reshape(-1, 3)
    unique_colors = np.unique(mask_flat, axis=0)

    for color in unique_colors:
        if np.array_equal(color, [0, 0, 0]):
            indices = np.where(np.all(input_mask == color, axis=-1))
            output_mask[indices] = np.array([255, 255, 255, 0], dtype=np.uint8)
        else:
            indices = np.where(np.all(input_mask == color, axis=-1))
            output_mask[indices] = np.array(list(color) + [255], dtype=np.uint8)

    io.imsave(output_mask_path, output_mask)

4. Prepare JSON File

Prepare the JSON to import to Hub, in the following format:

Where externalId is the external ID of the asset to pre-annotate, brushDataUrl is a link to your publicly accessible PNG mask, and tools is the list of instances available in the mask with randomly assigned unique RGB color information ("brush") and the "schemaId" of each class.

For Brush Traces with Publicly Accessible URLs

[
  {
    "externalId": "0a6e52c9-1eda-426f-8bcb-e6b1cd45668c.png",
    "brushDataUrl": "https://domain.s3.eu-central-1.amazonaws.com/brush-import-samples/0a6e52c9-1eda-426f-8bcb-e6b1cd45668c.png",
    "tools": [{"brush": [244, 67, 54], "schemaId": "29fde0dbe3eca1115f18368"}]
  },
  {
    "externalId": "0e9ee946-ebaa-4847-99ee-ce7386795333.png",
    "brushDataUrl": "https://domain.s3.eu-central-1.amazonaws.com/brush-import-samples/0e9ee946-ebaa-4847-99ee-ce7386795333.png",
    "tools": [{"brush": [244, 67, 54], "schemaId": "29fde0dbe3eca1115f18368"}]
  },
  {
    "externalId": "0eaf2333-5533-4353-ac3d-55774b6adb47.png",
    "brushDataUrl": "https://domain.s3.eu-central-1.amazonaws.com/brush-import-samples/0eaf2333-5533-4353-ac3d-55774b6adb47.png",
    "tools": [{"brush": [244, 67, 54], "schemaId": "29fde0dbe3eca1115f18368"}]
  },
  {
    "externalId": "0ebf0551-6c2d-4b6e-9a46-d5e1a5f22e72.png",
    "brushDataUrl": "https://domain.s3.eu-central-1.amazonaws.com/brush-import-samples/0ebf0551-6c2d-4b6e-9a46-d5e1a5f22e72.png",
    "tools": [{"brush": [244, 67, 54], "schemaId": "29fde0dbe3eca1115f18368"}]
  }
]

If your masks are stored in a cloud storage service, ensure the bucket's CORS settings are set properly, or your masks will not be visible on Hub.

Here is a guide on how to set up CORS in such a way that your masks will be visible.

For Brush Traces in Private Buckets

• Integrate your private bucket with Ango Hub by following the instructions on the Storages docs page.
• Navigate to your organization's Organization page, then navigate to the Storages tab.

• Copy the ID of the storage where your brush traces are stored.

• Prepare a JSON with the following format:

[
  {
    "externalId": "0a6e52c9-1eda-426f-8bcb-e6b1cd45668c.png",
    "brushDataUrl": "https://url.com/e6b1cd45668c.png?storageId=<YOUR_STORAGE_ID>",
    "tools": [{"brush": [244, 67, 54], "schemaId": "29fde0dbe3eca1115f18368"}]
  },
  {
    "externalId": "0e9ee946-ebaa-4847-99ee-ce7386795333.png",
    "brushDataUrl": "https://url.com/ce7386795333.png?storageId=<YOUR_STORAGE_ID>",
    "tools": [{"brush": [244, 67, 54], "schemaId": "29fde0dbe3eca1115f18368"}]
  }
]

5. Import Annotations

In your project, navigate to the Assets tab and ensure that the assets you wish to prelabel are in the Start stage, then click on Import. Drag and drop the JSON file you have just prepared. Your assets will be pre-labeled.

To programmatically import brush traces, you can use the import_labels function available in our SDK.