cn.smartjavaai.face.model.facedect.mtcnn

类 PNetModel

java.lang.Object

cn.smartjavaai.face.model.facedect.mtcnn.PNetModel

public class PNetModel
extends Object

作者:

dwj

构造器概要

构造器

构造器和说明
PNetModel()

方法概要

限定符和类型	方法和说明
static ai.djl.ndarray.NDList	firstStage(ai.djl.ndarray.NDManager manager, ai.djl.inference.Predictor<ai.djl.ndarray.NDList,ai.djl.ndarray.NDList> pnetPredictor, ai.djl.modality.cv.Image image)
static List<ai.djl.ndarray.NDArray>	generateBoundingBox(ai.djl.ndarray.NDArray reg, ai.djl.ndarray.NDArray probs, float scale, float threshold) 生成候选框，等价于 Python 版 generateBoundingBox
static List<Double>	generateScales(ai.djl.modality.cv.Image image) 生成金字塔缩放比例列表
static ai.djl.ndarray.NDArray	pNetPre(ai.djl.modality.cv.Image input, ai.djl.ndarray.NDManager manager)

从类继承的方法 java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

构造器详细资料

PNetModel

public PNetModel()

方法详细资料

generateScales

public static List<Double> generateScales(ai.djl.modality.cv.Image image)

生成金字塔缩放比例列表

参数:

image -

返回:

pNetPre

public static ai.djl.ndarray.NDArray pNetPre(ai.djl.modality.cv.Image input,
                                             ai.djl.ndarray.NDManager manager)

firstStage

public static ai.djl.ndarray.NDList firstStage(ai.djl.ndarray.NDManager manager,
                                               ai.djl.inference.Predictor<ai.djl.ndarray.NDList,ai.djl.ndarray.NDList> pnetPredictor,
                                               ai.djl.modality.cv.Image image)
                                        throws ai.djl.translate.TranslateException

抛出:

ai.djl.translate.TranslateException

generateBoundingBox

public static List<ai.djl.ndarray.NDArray> generateBoundingBox(ai.djl.ndarray.NDArray reg,
                                                               ai.djl.ndarray.NDArray probs,
                                                               float scale,
                                                               float threshold)

生成候选框，等价于 Python 版 generateBoundingBox

参数:

reg - NDArray [B,4,H,W]，回归偏移量

probs - NDArray [B,H,W]，人脸概率

scale - 当前金字塔缩放比例

threshold - 阈值

返回:

一个包含两个元素的 List： 0 -> NDArray bounding boxes [N,9] (x1,y1,x2,y2,score,dx1,dy1,dx2,dy2) 1 -> NDArray image_inds [N]