ml.postprocessing.mediapipe — Google Mediapipe

The ml.postprocessing.mediapipe module contains classes for Google Mediapipe.

class BlazeFace – Face Detection

Used to post-process BlazeFace model output.

Constructors

class ml.postprocessing.mediapipe.BlazeFace(threshold: float = 0.6, anchors=None, nms_threshold: float = 0.1, nms_sigma: float = 0.1) → BlazeFace

Create a BlazeFace postprocessor.

threshold The threshold to use for postprocessing.

anchors A list of anchor points (x, y) the model was trained on to use for postprocessing.

This post-processor returns a list of rect [x, y, w, h], score, and (x, y) keypoint list tuples. E.g. [((x, y, w, h), score, keypoints)]. Note that empty class list are included in the output to ensure the position of each class list in the output matches the position of the class index in the model output.

class BlazePalm – Palm Detection

Used to post-process BlazePalm model output.

Constructors

class ml.postprocessing.mediapipe.BlazePalm(threshold: float = 0.6, anchors=None, nms_threshold: float = 0.1, nms_sigma: float = 0.1) → BlazePalm

Create a BlazePalm postprocessor.

threshold The threshold to use for postprocessing.

anchors A list of anchor points (x, y) the model was trained on to use for postprocessing.

This post-processor returns a list of rect [x, y, w, h], score, and (x, y) keypoint list tuples. E.g. [((x, y, w, h), score, keypoints)]. Note that empty class list are included in the output to ensure the position of each class list in the output matches the position of the class index in the model output.

class FaceLandmarks – Face Landmarks

Used to post-process FaceLandmarks model output.

Constructors

class ml.postprocessing.mediapipe.FaceLandmarks(threshold: float = 0.6, nms_threshold: float = 0.1, nms_sigma: float = 0.1) → FaceLandmarks

Create a FaceLandmarks postprocessor.

threshold The threshold to use for postprocessing.

This post-processor returns a list of rect [x, y, w, h], score, and (x, y, z) keypoint list tuples. E.g. [((x, y, w, h), score, keypoints)]. Note that empty class list are included in the output to ensure the position of each class list in the output matches the position of the class index in the model output.

class HandLandmarks – Hand Landmarks

Used to post-process HandLandmarks model output.

Constructors

class ml.postprocessing.mediapipe.HandLandmarks(threshold: float = 0.6, nms_threshold: float = 0.1, nms_sigma: float = 0.1) → HandLandmarks

Create a HandLandmarks postprocessor.

threshold The threshold to use for postprocessing.

This post-processor returns a list of rect [x, y, w, h], score, and (x, y, z) keypoint list tuples for each class in the model output (left=0, or right=1) E.g. [[((x, y, w, h), score, keypoints)]]. Note that empty class list are included in the output to ensure the position of each class list in the output matches the position of the class index in the model output.