mtcnn_detector.py

# coding: utf-8
import os
import mxnet as mx
import numpy as np
import math
import cv2
from multiprocessing import Pool
from itertools import repeat
#from itertools import izip
izip = zip
from helper import nms, adjust_input, generate_bbox, detect_first_stage_warpper

class MtcnnDetector(object):
    """
        Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Neural Networks
        see https://github.com/kpzhang93/MTCNN_face_detection_alignment
        this is a mxnet version
    """
    def __init__(self,
                 model_folder='.',
                 minsize = 20,
                 threshold = [0.6, 0.7, 0.8],
                 factor = 0.709,
                 num_worker = 1,
                 accurate_landmark = False,
                 ctx=mx.cpu()):
        """
            Initialize the detector

            Parameters:
            ----------
                model_folder : string
                    path for the models
                minsize : float number
                    minimal face to detect
                threshold : float number
                    detect threshold for 3 stages
                factor: float number
                    scale factor for image pyramid
                num_worker: int number
                    number of processes we use for first stage
                accurate_landmark: bool
                    use accurate landmark localization or not

        """
        self.num_worker = num_worker
        self.accurate_landmark = accurate_landmark

        # load 4 models from folder
        models = ['det1', 'det2', 'det3','det4']
        models = [ os.path.join(model_folder, f) for f in models]
        
        self.PNets = []
        for i in range(num_worker):
            workner_net = mx.model.FeedForward.load(models[0], 1, ctx=ctx)
            self.PNets.append(workner_net)

        #self.Pool = Pool(num_worker)

        self.RNet = mx.model.FeedForward.load(models[1], 1, ctx=ctx)
        self.ONet = mx.model.FeedForward.load(models[2], 1, ctx=ctx)
        self.LNet = mx.model.FeedForward.load(models[3], 1, ctx=ctx)

        self.minsize   = float(minsize)
        self.factor    = float(factor)
        self.threshold = threshold


    def convert_to_square(self, bbox):
        """
            convert bbox to square

        Parameters:
        ----------
            bbox: numpy array , shape n x 5
                input bbox

        Returns:
        -------
            square bbox
        """
        square_bbox = bbox.copy()

        h = bbox[:, 3] - bbox[:, 1] + 1
        w = bbox[:, 2] - bbox[:, 0] + 1
        max_side = np.maximum(h,w)
        square_bbox[:, 0] = bbox[:, 0] + w*0.5 - max_side*0.5
        square_bbox[:, 1] = bbox[:, 1] + h*0.5 - max_side*0.5
        square_bbox[:, 2] = square_bbox[:, 0] + max_side - 1
        square_bbox[:, 3] = square_bbox[:, 1] + max_side - 1
        return square_bbox

    def calibrate_box(self, bbox, reg):
        """
            calibrate bboxes

        Parameters:
        ----------
            bbox: numpy array, shape n x 5
                input bboxes
            reg:  numpy array, shape n x 4
                bboxex adjustment

        Returns:
        -------
            bboxes after refinement

        """
        w = bbox[:, 2] - bbox[:, 0] + 1
        w = np.expand_dims(w, 1)
        h = bbox[:, 3] - bbox[:, 1] + 1
        h = np.expand_dims(h, 1)
        reg_m = np.hstack([w, h, w, h])
        aug = reg_m * reg
        bbox[:, 0:4] = bbox[:, 0:4] + aug
        return bbox

 
    def pad(self, bboxes, w, h):
        """
            pad the the bboxes, alse restrict the size of it

        Parameters:
        ----------
            bboxes: numpy array, n x 5
                input bboxes
            w: float number
                width of the input image
            h: float number
                height of the input image
        Returns :
        ------s
            dy, dx : numpy array, n x 1
                start point of the bbox in target image
            edy, edx : numpy array, n x 1
                end point of the bbox in target image
            y, x : numpy array, n x 1
                start point of the bbox in original image
            ex, ex : numpy array, n x 1
                end point of the bbox in original image
            tmph, tmpw: numpy array, n x 1
                height and width of the bbox

        """
        tmpw, tmph = bboxes[:, 2] - bboxes[:, 0] + 1,  bboxes[:, 3] - bboxes[:, 1] + 1
        num_box = bboxes.shape[0]

        dx , dy= np.zeros((num_box, )), np.zeros((num_box, ))
        edx, edy  = tmpw.copy()-1, tmph.copy()-1

        x, y, ex, ey = bboxes[:, 0], bboxes[:, 1], bboxes[:, 2], bboxes[:, 3]

        tmp_index = np.where(ex > w-1)
        edx[tmp_index] = tmpw[tmp_index] + w - 2 - ex[tmp_index]
        ex[tmp_index] = w - 1

        tmp_index = np.where(ey > h-1)
        edy[tmp_index] = tmph[tmp_index] + h - 2 - ey[tmp_index]
        ey[tmp_index] = h - 1

        tmp_index = np.where(x < 0)
        dx[tmp_index] = 0 - x[tmp_index]
        x[tmp_index] = 0

        tmp_index = np.where(y < 0)
        dy[tmp_index] = 0 - y[tmp_index]
        y[tmp_index] = 0

        return_list = [dy, edy, dx, edx, y, ey, x, ex, tmpw, tmph]
        return_list = [item.astype(np.int32) for item in return_list]

        return  return_list

    def slice_index(self, number):
        """
            slice the index into (n,n,m), m < n
        Parameters:
        ----------
            number: int number
                number
        """
        def chunks(l, n):
            """Yield successive n-sized chunks from l."""
            for i in range(0, len(l), n):
                yield l[i:i + n]
        num_list = range(number)
        return list(chunks(num_list, self.num_worker))
        
    def detect_face_limited(self, img, det_type=2):
        height, width, _ = img.shape
        if det_type>=2:
          total_boxes = np.array( [ [0.0, 0.0, img.shape[1], img.shape[0], 0.9] ] ,dtype=np.float32)
          num_box = total_boxes.shape[0]

          # pad the bbox
          [dy, edy, dx, edx, y, ey, x, ex, tmpw, tmph] = self.pad(total_boxes, width, height)
          # (3, 24, 24) is the input shape for RNet
          input_buf = np.zeros((num_box, 3, 24, 24), dtype=np.float32)

          for i in range(num_box):
              tmp = np.zeros((tmph[i], tmpw[i], 3), dtype=np.uint8)
              tmp[dy[i]:edy[i]+1, dx[i]:edx[i]+1, :] = img[y[i]:ey[i]+1, x[i]:ex[i]+1, :]
              input_buf[i, :, :, :] = adjust_input(cv2.resize(tmp, (24, 24)))

          output = self.RNet.predict(input_buf)

          # filter the total_boxes with threshold
          passed = np.where(output[1][:, 1] > self.threshold[1])
          total_boxes = total_boxes[passed]

          if total_boxes.size == 0:
              return None

          total_boxes[:, 4] = output[1][passed, 1].reshape((-1,))
          reg = output[0][passed]

          # nms
          pick = nms(total_boxes, 0.7, 'Union')
          total_boxes = total_boxes[pick]
          total_boxes = self.calibrate_box(total_boxes, reg[pick])
          total_boxes = self.convert_to_square(total_boxes)
          total_boxes[:, 0:4] = np.round(total_boxes[:, 0:4])
        else:
          total_boxes = np.array( [ [0.0, 0.0, img.shape[1], img.shape[0], 0.9] ] ,dtype=np.float32)
        num_box = total_boxes.shape[0]
        [dy, edy, dx, edx, y, ey, x, ex, tmpw, tmph] = self.pad(total_boxes, width, height)
        # (3, 48, 48) is the input shape for ONet
        input_buf = np.zeros((num_box, 3, 48, 48), dtype=np.float32)

        for i in range(num_box):
            tmp = np.zeros((tmph[i], tmpw[i], 3), dtype=np.float32)
            tmp[dy[i]:edy[i]+1, dx[i]:edx[i]+1, :] = img[y[i]:ey[i]+1, x[i]:ex[i]+1, :]
            input_buf[i, :, :, :] = adjust_input(cv2.resize(tmp, (48, 48)))

        output = self.ONet.predict(input_buf)
        #print(output[2])

        # filter the total_boxes with threshold
        passed = np.where(output[2][:, 1] > self.threshold[2])
        total_boxes = total_boxes[passed]

        if total_boxes.size == 0:
            return None

        total_boxes[:, 4] = output[2][passed, 1].reshape((-1,))
        reg = output[1][passed]
        points = output[0][passed]

        # compute landmark points
        bbw = total_boxes[:, 2] - total_boxes[:, 0] + 1
        bbh = total_boxes[:, 3] - total_boxes[:, 1] + 1
        points[:, 0:5] = np.expand_dims(total_boxes[:, 0], 1) + np.expand_dims(bbw, 1) * points[:, 0:5]
        points[:, 5:10] = np.expand_dims(total_boxes[:, 1], 1) + np.expand_dims(bbh, 1) * points[:, 5:10]

        # nms
        total_boxes = self.calibrate_box(total_boxes, reg)
        pick = nms(total_boxes, 0.7, 'Min')
        total_boxes = total_boxes[pick]
        points = points[pick]
        
        if not self.accurate_landmark:
            return total_boxes, points

        #############################################
        # extended stage
        #############################################
        num_box = total_boxes.shape[0]
        patchw = np.maximum(total_boxes[:, 2]-total_boxes[:, 0]+1, total_boxes[:, 3]-total_boxes[:, 1]+1)
        patchw = np.round(patchw*0.25)

        # make it even
        patchw[np.where(np.mod(patchw,2) == 1)] += 1

        input_buf = np.zeros((num_box, 15, 24, 24), dtype=np.float32)
        for i in range(5):
            x, y = points[:, i], points[:, i+5]
            x, y = np.round(x-0.5*patchw), np.round(y-0.5*patchw)
            [dy, edy, dx, edx, y, ey, x, ex, tmpw, tmph] = self.pad(np.vstack([x, y, x+patchw-1, y+patchw-1]).T,
                                                                    width,
                                                                    height)
            for j in range(num_box):
                tmpim = np.zeros((tmpw[j], tmpw[j], 3), dtype=np.float32)
                tmpim[dy[j]:edy[j]+1, dx[j]:edx[j]+1, :] = img[y[j]:ey[j]+1, x[j]:ex[j]+1, :]
                input_buf[j, i*3:i*3+3, :, :] = adjust_input(cv2.resize(tmpim, (24, 24)))

        output = self.LNet.predict(input_buf)

        pointx = np.zeros((num_box, 5))
        pointy = np.zeros((num_box, 5))

        for k in range(5):
            # do not make a large movement
            tmp_index = np.where(np.abs(output[k]-0.5) > 0.35)
            output[k][tmp_index[0]] = 0.5

            pointx[:, k] = np.round(points[:, k] - 0.5*patchw) + output[k][:, 0]*patchw
            pointy[:, k] = np.round(points[:, k+5] - 0.5*patchw) + output[k][:, 1]*patchw

        points = np.hstack([pointx, pointy])
        points = points.astype(np.int32)

        return total_boxes, points

    def detect_face(self, img, det_type=0):
        """
            detect face over img
        Parameters:
        ----------
            img: numpy array, bgr order of shape (1, 3, n, m)
                input image
        Retures:
        -------
            bboxes: numpy array, n x 5 (x1,y2,x2,y2,score)
                bboxes
            points: numpy array, n x 10 (x1, x2 ... x5, y1, y2 ..y5)
                landmarks
        """

        # check input
        height, width, _ = img.shape
        if det_type==0:
            MIN_DET_SIZE = 12

            if img is None:
                return None

            # only works for color image
            if len(img.shape) != 3:
                return None

            # detected boxes
            total_boxes = []

            minl = min( height, width)

            # get all the valid scales
            scales = []
            m = MIN_DET_SIZE/self.minsize
            minl *= m
            factor_count = 0
            while minl > MIN_DET_SIZE:
                scales.append(m*self.factor**factor_count)
                minl *= self.factor
                factor_count += 1

            #############################################
            # first stage
            #############################################
            #for scale in scales:
            #    return_boxes = self.detect_first_stage(img, scale, 0)
            #    if return_boxes is not None:
            #        total_boxes.append(return_boxes)
            
            sliced_index = self.slice_index(len(scales))
            total_boxes = []
            for batch in sliced_index:
                #local_boxes = self.Pool.map( detect_first_stage_warpper, \
                #        izip(repeat(img), self.PNets[:len(batch)], [scales[i] for i in batch], repeat(self.threshold[0])) )
                local_boxes = map( detect_first_stage_warpper, \
                        izip(repeat(img), self.PNets[:len(batch)], [scales[i] for i in batch], repeat(self.threshold[0])) )
                total_boxes.extend(local_boxes)
            
            # remove the Nones 
            total_boxes = [ i for i in total_boxes if i is not None]

            if len(total_boxes) == 0:
                return None
            
            total_boxes = np.vstack(total_boxes)

            if total_boxes.size == 0:
                return None

            # merge the detection from first stage
            pick = nms(total_boxes[:, 0:5], 0.7, 'Union')
            total_boxes = total_boxes[pick]

            bbw = total_boxes[:, 2] - total_boxes[:, 0] + 1
            bbh = total_boxes[:, 3] - total_boxes[:, 1] + 1

            # refine the bboxes
            total_boxes = np.vstack([total_boxes[:, 0]+total_boxes[:, 5] * bbw,
                                     total_boxes[:, 1]+total_boxes[:, 6] * bbh,
                                     total_boxes[:, 2]+total_boxes[:, 7] * bbw,
                                     total_boxes[:, 3]+total_boxes[:, 8] * bbh,
                                     total_boxes[:, 4]
                                     ])

            total_boxes = total_boxes.T
            total_boxes = self.convert_to_square(total_boxes)
            total_boxes[:, 0:4] = np.round(total_boxes[:, 0:4])
        else:
            total_boxes = np.array( [ [0.0, 0.0, img.shape[1], img.shape[0], 0.9] ] ,dtype=np.float32)

        #############################################
        # second stage
        #############################################
        num_box = total_boxes.shape[0]

        # pad the bbox
        [dy, edy, dx, edx, y, ey, x, ex, tmpw, tmph] = self.pad(total_boxes, width, height)
        # (3, 24, 24) is the input shape for RNet
        input_buf = np.zeros((num_box, 3, 24, 24), dtype=np.float32)

        for i in range(num_box):
            tmp = np.zeros((tmph[i], tmpw[i], 3), dtype=np.uint8)
            tmp[dy[i]:edy[i]+1, dx[i]:edx[i]+1, :] = img[y[i]:ey[i]+1, x[i]:ex[i]+1, :]
            input_buf[i, :, :, :] = adjust_input(cv2.resize(tmp, (24, 24)))

        output = self.RNet.predict(input_buf)

        # filter the total_boxes with threshold
        passed = np.where(output[1][:, 1] > self.threshold[1])
        total_boxes = total_boxes[passed]

        if total_boxes.size == 0:
            return None

        total_boxes[:, 4] = output[1][passed, 1].reshape((-1,))
        reg = output[0][passed]

        # nms
        pick = nms(total_boxes, 0.7, 'Union')
        total_boxes = total_boxes[pick]
        total_boxes = self.calibrate_box(total_boxes, reg[pick])
        total_boxes = self.convert_to_square(total_boxes)
        total_boxes[:, 0:4] = np.round(total_boxes[:, 0:4])

        #############################################
        # third stage
        #############################################
        num_box = total_boxes.shape[0]

        # pad the bbox
        [dy, edy, dx, edx, y, ey, x, ex, tmpw, tmph] = self.pad(total_boxes, width, height)
        # (3, 48, 48) is the input shape for ONet
        input_buf = np.zeros((num_box, 3, 48, 48), dtype=np.float32)

        for i in range(num_box):
            tmp = np.zeros((tmph[i], tmpw[i], 3), dtype=np.float32)
            tmp[dy[i]:edy[i]+1, dx[i]:edx[i]+1, :] = img[y[i]:ey[i]+1, x[i]:ex[i]+1, :]
            input_buf[i, :, :, :] = adjust_input(cv2.resize(tmp, (48, 48)))

        output = self.ONet.predict(input_buf)

        # filter the total_boxes with threshold
        passed = np.where(output[2][:, 1] > self.threshold[2])
        total_boxes = total_boxes[passed]

        if total_boxes.size == 0:
            return None

        total_boxes[:, 4] = output[2][passed, 1].reshape((-1,))
        reg = output[1][passed]
        points = output[0][passed]

        # compute landmark points
        bbw = total_boxes[:, 2] - total_boxes[:, 0] + 1
        bbh = total_boxes[:, 3] - total_boxes[:, 1] + 1
        points[:, 0:5] = np.expand_dims(total_boxes[:, 0], 1) + np.expand_dims(bbw, 1) * points[:, 0:5]
        points[:, 5:10] = np.expand_dims(total_boxes[:, 1], 1) + np.expand_dims(bbh, 1) * points[:, 5:10]

        # nms
        total_boxes = self.calibrate_box(total_boxes, reg)
        pick = nms(total_boxes, 0.7, 'Min')
        total_boxes = total_boxes[pick]
        points = points[pick]
        
        if not self.accurate_landmark:
            return total_boxes, points

        #############################################
        # extended stage
        #############################################
        num_box = total_boxes.shape[0]
        patchw = np.maximum(total_boxes[:, 2]-total_boxes[:, 0]+1, total_boxes[:, 3]-total_boxes[:, 1]+1)
        patchw = np.round(patchw*0.25)

        # make it even
        patchw[np.where(np.mod(patchw,2) == 1)] += 1

        input_buf = np.zeros((num_box, 15, 24, 24), dtype=np.float32)
        for i in range(5):
            x, y = points[:, i], points[:, i+5]
            x, y = np.round(x-0.5*patchw), np.round(y-0.5*patchw)
            [dy, edy, dx, edx, y, ey, x, ex, tmpw, tmph] = self.pad(np.vstack([x, y, x+patchw-1, y+patchw-1]).T,
                                                                    width,
                                                                    height)
            for j in range(num_box):
                tmpim = np.zeros((tmpw[j], tmpw[j], 3), dtype=np.float32)
                tmpim[dy[j]:edy[j]+1, dx[j]:edx[j]+1, :] = img[y[j]:ey[j]+1, x[j]:ex[j]+1, :]
                input_buf[j, i*3:i*3+3, :, :] = adjust_input(cv2.resize(tmpim, (24, 24)))

        output = self.LNet.predict(input_buf)

        pointx = np.zeros((num_box, 5))
        pointy = np.zeros((num_box, 5))

        for k in range(5):
            # do not make a large movement
            tmp_index = np.where(np.abs(output[k]-0.5) > 0.35)
            output[k][tmp_index[0]] = 0.5

            pointx[:, k] = np.round(points[:, k] - 0.5*patchw) + output[k][:, 0]*patchw
            pointy[:, k] = np.round(points[:, k+5] - 0.5*patchw) + output[k][:, 1]*patchw

        points = np.hstack([pointx, pointy])
        points = points.astype(np.int32)

        return total_boxes, points



    def list2colmatrix(self, pts_list):
        """
            convert list to column matrix
        Parameters:
        ----------
            pts_list:
                input list
        Retures:
        -------
            colMat: 

        """
        assert len(pts_list) > 0
        colMat = []
        for i in range(len(pts_list)):
            colMat.append(pts_list[i][0])
            colMat.append(pts_list[i][1])
        colMat = np.matrix(colMat).transpose()
        return colMat

    def find_tfrom_between_shapes(self, from_shape, to_shape):
        """
            find transform between shapes
        Parameters:
        ----------
            from_shape: 
            to_shape: 
        Retures:
        -------
            tran_m:
            tran_b:
        """
        assert from_shape.shape[0] == to_shape.shape[0] and from_shape.shape[0] % 2 == 0

        sigma_from = 0.0
        sigma_to = 0.0
        cov = np.matrix([[0.0, 0.0], [0.0, 0.0]])

        # compute the mean and cov
        from_shape_points = from_shape.reshape(from_shape.shape[0]/2, 2)
        to_shape_points = to_shape.reshape(to_shape.shape[0]/2, 2)
        mean_from = from_shape_points.mean(axis=0)
        mean_to = to_shape_points.mean(axis=0)

        for i in range(from_shape_points.shape[0]):
            temp_dis = np.linalg.norm(from_shape_points[i] - mean_from)
            sigma_from += temp_dis * temp_dis
            temp_dis = np.linalg.norm(to_shape_points[i] - mean_to)
            sigma_to += temp_dis * temp_dis
            cov += (to_shape_points[i].transpose() - mean_to.transpose()) * (from_shape_points[i] - mean_from)

        sigma_from = sigma_from / to_shape_points.shape[0]
        sigma_to = sigma_to / to_shape_points.shape[0]
        cov = cov / to_shape_points.shape[0]

        # compute the affine matrix
        s = np.matrix([[1.0, 0.0], [0.0, 1.0]])
        u, d, vt = np.linalg.svd(cov)

        if np.linalg.det(cov) < 0:
            if d[1] < d[0]:
                s[1, 1] = -1
            else:
                s[0, 0] = -1
        r = u * s * vt
        c = 1.0
        if sigma_from != 0:
            c = 1.0 / sigma_from * np.trace(np.diag(d) * s)

        tran_b = mean_to.transpose() - c * r * mean_from.transpose()
        tran_m = c * r

        return tran_m, tran_b

    def extract_image_chips(self, img, points, desired_size=256, padding=0):
        """
            crop and align face
        Parameters:
        ----------
            img: numpy array, bgr order of shape (1, 3, n, m)
                input image
            points: numpy array, n x 10 (x1, x2 ... x5, y1, y2 ..y5)
            desired_size: default 256
            padding: default 0
        Retures:
        -------
            crop_imgs: list, n
                cropped and aligned faces 
        """
        crop_imgs = []
        for p in points:
            shape  =[]
            for k in range(len(p)/2):
                shape.append(p[k])
                shape.append(p[k+5])

            if padding > 0:
                padding = padding
            else:
                padding = 0
            # average positions of face points
            mean_face_shape_x = [0.224152, 0.75610125, 0.490127, 0.254149, 0.726104]
            mean_face_shape_y = [0.2119465, 0.2119465, 0.628106, 0.780233, 0.780233]

            from_points = []
            to_points = []

            for i in range(len(shape)/2):
                x = (padding + mean_face_shape_x[i]) / (2 * padding + 1) * desired_size
                y = (padding + mean_face_shape_y[i]) / (2 * padding + 1) * desired_size
                to_points.append([x, y])
                from_points.append([shape[2*i], shape[2*i+1]])

            # convert the points to Mat
            from_mat = self.list2colmatrix(from_points)
            to_mat = self.list2colmatrix(to_points)

            # compute the similar transfrom
            tran_m, tran_b = self.find_tfrom_between_shapes(from_mat, to_mat)

            probe_vec = np.matrix([1.0, 0.0]).transpose()
            probe_vec = tran_m * probe_vec

            scale = np.linalg.norm(probe_vec)
            angle = 180.0 / math.pi * math.atan2(probe_vec[1, 0], probe_vec[0, 0])

            from_center = [(shape[0]+shape[2])/2.0, (shape[1]+shape[3])/2.0]
            to_center = [0, 0]
            to_center[1] = desired_size * 0.4
            to_center[0] = desired_size * 0.5

            ex = to_center[0] - from_center[0]
            ey = to_center[1] - from_center[1]

            rot_mat = cv2.getRotationMatrix2D((from_center[0], from_center[1]), -1*angle, scale)
            rot_mat[0][2] += ex
            rot_mat[1][2] += ey

            chips = cv2.warpAffine(img, rot_mat, (desired_size, desired_size))
            crop_imgs.append(chips)

        return crop_imgs