1. Request access through the model page on Hugging Face and wait for approval from the model authors.
2. Create a Custom keys workspace connection in Microsoft Foundry or Azure Machine Learning named HuggingFaceTokenConnection with the key HF_TOKEN and value your Hugging Face read or fine-grained token (marked as secret).
3. Create the Managed Online Endpoint with the property enforce_access_to_default_secret_stores set to enabled so it can access the secret connection value.
4. Once access is approved, the connection is configured, and the endpoint is created with read access to the token, you can deploy and use the model in Microsoft Foundry or Azure Machine Learning.

1. Request access through the model page on Hugging Face and wait for approval from the model authors.
2. Create a Custom keys workspace connection in Microsoft Foundry or Azure Machine Learning named HuggingFaceTokenConnection with the key HF_TOKEN and value your Hugging Face read or fine-grained token (marked as secret).
3. Create the Managed Online Endpoint with the property enforce_access_to_default_secret_stores set to enabled so it can access the secret connection value.
4. Once access is approved, the connection is configured, and the endpoint is created with read access to the token, you can deploy and use the model in Microsoft Foundry or Azure Machine Learning.

• Original Model Card
• mask-generation Task on Hugging Face

Mask Generation

Send Request

curl <AZUREML_ENDPOINT_URL> \
    -X POST \
    -H "Authorization: Bearer <AZUREML_TOKEN>" \
    -H "Content-Type: application/json" \
    -d '{"inputs":"http://images.cocodataset.org/val2017/000000077595.jpg","parameters":{"mask_threshold":0.5}}'

Supported Parameters

• inputs (string or bytes): The input image, which can be a URL, a local file path, or raw bytes representing an image.
• parameters (object, optional):
  • mask_threshold (float, optional): Threshold applied to the predicted masks to convert them into binary values. Defaults to 0.0.
  • pred_iou_thresh (float, optional): Filtering threshold in [0, 1] applied to the predicted mask IoU/quality scores. Defaults to 0.88.
  • stability_score_thresh (float, optional): Filtering threshold in [0, 1] using the stability of the mask under variations of the binarization cutoff. Defaults to 0.95.
  • stability_score_offset (integer, optional): Amount by which the cutoff is shifted when computing the stability score. Defaults to 1.
  • crops_nms_thresh (float, optional): Box IoU cutoff used by non-maximum suppression to remove duplicate masks across crops. Defaults to 7.
  • crops_n_layers (integer, optional): Number of crop layers; if greater than 0, the image is recursively cropped into (2^{\text{layer}}) regions per layer for additional mask prediction passes. Defaults to 0.
  • crop_overlap_ratio (float, optional): Fraction of the image side length by which crops overlap in the first crop layer, scaled accordingly for deeper layers. Defaults to 512 / 1500 i.e., 0.3413.
  • crop_n_points_downscale_factor (integer, optional): Factor controlling how the number of points-per-side decreases with each crop layer, using (\text{factor}^n) at layer (n). Defaults to 1.
  • timeout (float, optional): Maximum time in seconds to wait when fetching remote images; if null, no timeout is enforced and the call may block indefinitely.

Expected output

• mask (str): The generated mask over the original image encoded in base64. If the image is blank, it means that no objects have been found.
• score (float, optional): Optionally, when the model is capable of estimating a confidence of the detected object described by the mask.

Promptable Concept Segmentation (PCS)

Send Request

curl <AZUREML_ENDPOINT_URL>/promptable-concept-segmentation \
    -X POST \
    -H "Authorization: Bearer <AZUREML_TOKEN>" \
    -H "Content-Type: application/json" \
    -d '{"inputs":{"image":"http://images.cocodataset.org/val2017/000000077595.jpg","text":"cat"},"parameters":{"mask_threshold":0.5}}'

Supported Parameters

• inputs (object):
  • image (string or bytes): The input image, which can be a URL, a local file path, or raw bytes representing an image.
  • text (string): The concept to find and segment within the provided image.
• parameters (object, optional):
  • threshold (float, optional): Score threshold to keep instance predictions. Defaults to 0.3.
  • mask_threshold (float, optional): Threshold for binarizing the predicted masks. Defaults to 0.5.

Expected output

• mask (str): The generated mask over the original image encoded in base64. If the image is blank, it means that no objects have been found.
• score (float, optional): Optionally, when the model is capable of estimating a confidence of the detected object described by the mask.
• box ([float, float, float, float], optional): Optionally, when the model is capable of generating the bounding boxes in absolute pixel coordinates (xyxy format).