Supported content types: image/jpeg

This model accepts images in the mime-type specified above.

The image must be at least 224x224. The model resizes the image to the size specified by the user before performing the inference. Better results are achieved with 16:9 image proportions.

Content type: text/json

For every image, the model returns a single JSON file with all the detections.

The model returns JSON object, that includes an array with individual elements for each face detected. Each element has three attributes:

1. box_points: includes the bounding box around the detected face. Each bounding box consists of four numbers in [X1 Y1 X2 Y2] format in the source image coordinates.
2. classes: “no_mask” represents the probability score that the face in this bounding box does not have a mask. Probability is given in percents (0..100 range)

Prediction method can take four additional parameters: face_detector_arch - [resnet18, mobilenet_v2] - the architecture of mask classifier subnetwork face_crop_size - [64, 128, 192, 384] - final size, to which extracted faces will be resized before classification (higher is slower) face_detector_image_height - height of the image to be resized internally (higher is slower) face_detector_confidence - confidence value (in 0..1 range) to filter boxes with faces Use higher values for images of better quality or bigger spatial resolution.

We recommend using this model for real-time inference for better utilization of the endpoint. Optionally, batch transform is also available.

*You can find more details here: https://github.com/VITechLab/aws-sagemaker-examples/tree/master/Medical-Mask-Detector *