Skip to content

affective_computing / mobile_face_recognize

Go to a project
Toggle navigation
Toggle navigation pinning
d77bff18 by jiangwenqiang
Options

remove q

1 parent f6b88746
Inline Side-by-side
Showing 1 changed file with 0 additions and 220 deletions
q deleted 100644 → 0
import os
import numpy as np
import MNN
import cv2
import logging
from retinaface import PriorBox
def py_cpu_nms(dets, thresh):
"""Pure Python NMS baseline."""
x1 = dets[:, 0]
y1 = dets[:, 1]
x2 = dets[:, 2]
y2 = dets[:, 3]
scores = dets[:, 4]
areas = (x2 - x1 + 1) * (y2 - y1 + 1)
order = scores.argsort()[::-1]
keep = []
while order.size > 0:
i = order[0]
keep.append(i)
xx1 = np.maximum(x1[i], x1[order[1:]])
yy1 = np.maximum(y1[i], y1[order[1:]])
xx2 = np.minimum(x2[i], x2[order[1:]])
yy2 = np.minimum(y2[i], y2[order[1:]])
w = np.maximum(0.0, xx2 - xx1 + 1)
h = np.maximum(0.0, yy2 - yy1 + 1)
inter = w * h
ovr = inter / (areas[i] + areas[order[1:]] - inter)
inds = np.where(ovr <= thresh)[0]
order = order[inds + 1]
return keep
def decode_landm(pre, priors, variances):
landms = np.concatenate((priors[:, :2] + pre[:, :2] * variances[0] * priors[:, 2:],
priors[:, :2] + pre[:, 2:4] * variances[0] * priors[:, 2:],
priors[:, :2] + pre[:, 4:6] * variances[0] * priors[:, 2:],
priors[:, :2] + pre[:, 6:8] * variances[0] * priors[:, 2:],
priors[:, :2] + pre[:, 8:10] * variances[0] * priors[:, 2:]), 1)
return landms
def decode(loc, priors, variances):
boxes = np.concatenate((
priors[:, :2] + loc[:, :2] * variances[0] * priors[:, 2:],
priors[:, 2:] * np.exp(loc[:, 2:] * variances[1])), 1)
boxes[:, :2] -= boxes[:, 2:] / 2
boxes[:, 2:] += boxes[:, :2]
return boxes
class Face_Detector(object):
def __init__(self, model_path):
logging.info('******** Start Init Face Detector ********')
self.det_interpreter = MNN.Interpreter(model_path)
self.det_session = self.det_interpreter.createSession()
self.det_input_tensor = self.det_interpreter.getSessionInput(self.det_session)
logging.info('******** Success Init Face Detector ********')
def detect(self, frame, thr):
logging.info('******** Start Face Detect ********')
input_size = 320
img = cv2.resize(frame, (input_size, input_size))
img = np.float32(img)
im_height, im_width, _ = img.shape
scale = np.array([img.shape[1], img.shape[0], img.shape[1], img.shape[0]])
scale1 = np.array([img.shape[1], img.shape[0], img.shape[1], img.shape[0],
img.shape[1], img.shape[0], img.shape[1], img.shape[0],
img.shape[1], img.shape[0]])
w_r = input_size/frame.shape[1]
h_r = input_size/frame.shape[0]
confidence_threshold = 0.02
vis_threshold = 0.5
nms_threshold = 0.4
keep_top_k = 100
variance = [0.1, 0.2]
img -= (104, 117, 123)
img = img.transpose(2, 0, 1)
img = np.expand_dims(img, axis=0)
input_tensor = MNN.Tensor((1, 3, input_size, input_size), MNN.Halide_Type_Float, img, MNN.Tensor_DimensionType_Caffe)
self.det_input_tensor.copyFrom(input_tensor)
self.det_interpreter.runSession(self.det_session)
bbox_output_tensor = self.det_interpreter.getSessionOutput(self.det_session, 'output0')
conf_output_tensor = self.det_interpreter.getSessionOutput(self.det_session, 'output1')
landmark_output_tensor = self.det_interpreter.getSessionOutput(self.det_session, 'output2')
bbox_output = bbox_output_tensor.getData()
conf_output = conf_output_tensor.getData()
landmark_output = landmark_output_tensor.getData()
norm_confs = list()
for i in range(int(len(conf_output)/2)):
norm_confs.append([conf_output[i * 2 + 0], conf_output[i * 2 + 1]])
norm_bboxes = list()
for i in range(int(len(conf_output)/2)):
norm_bboxes.append([bbox_output[i * 4 + 0], bbox_output[i * 4 + 1], bbox_output[i * 4 + 2], bbox_output[i * 4 + 3]])
norm_landmarks = list()
for i in range(int(len(conf_output)/2)):
norm_landmarks.append([landmark_output[i * 10 + 0], landmark_output[i * 10 + 1],
landmark_output[i * 10 + 2], landmark_output[i * 10 + 3],
landmark_output[i * 10 + 4], landmark_output[i * 10 + 5],
landmark_output[i * 10 + 6], landmark_output[i * 10 + 7],
landmark_output[i * 10 + 8], landmark_output[i * 10 + 9]])
norm_confs = np.array(norm_confs)
norm_bboxes = np.array(norm_bboxes)
norm_landmarks = np.array(norm_landmarks)
priorbox = PriorBox(image_size=(im_height, im_width))
priors = priorbox.forward()
scores = norm_confs[:, 1]
boxes = decode(norm_bboxes, priors, variance)
boxes = boxes * scale
landms = decode_landm(norm_landmarks, priors, variance)
landms = landms * scale1
# ignore low scores
inds = np.where(scores > confidence_threshold)[0]
boxes = boxes[inds]
landms = landms[inds]
scores = scores[inds]
# keep top-K before NMS
order = scores.argsort()[::-1][:keep_top_k]
boxes = boxes[order]
landms = landms[order]
scores = scores[order]
# do NMS
dets = np.hstack((boxes, scores[:, np.newaxis])).astype(np.float32, copy=False)
keep = py_cpu_nms(dets, nms_threshold)
dets = dets[keep, :]
landms = landms[keep]
# keep top-K faster NMS
dets = dets[:keep_top_k, :]
landms = landms[:keep_top_k, :]
dets = np.concatenate((dets, landms), axis=1)
face_bboxes = []
face_landmarks = []
max_area = float('-inf')
max_index = 0
i = 0
for b in dets:
if b[4] < vis_threshold:
continue
resize_b = []
x1 = int(b[0] / w_r)
y1 = int(b[1] / h_r)
x2 = int(b[2] / w_r)
y2 = int(b[3] / h_r)
x3 = int(b[5] / w_r)
y3 = int(b[6] / h_r)
x4 = int(b[7] / w_r)
y4 = int(b[8] / h_r)
x5 = int(b[9] / w_r)
y5 = int(b[10] / h_r)
x6 = int(b[11] / w_r)
y6 = int(b[12] / h_r)
x7 = int(b[13] / w_r)
y7 = int(b[14] / h_r)
resize_b = [x1, y1, x2, y2, 0, x3, y3, x4, y4, x5, y5, x6, y6, x7, y7]
# cv2.rectangle(frame, (resize_b[0], resize_b[1]), (resize_b[2], resize_b[3]), (0, 0, 255), 2)
# cv2.circle(frame, (resize_b[5], resize_b[6]), 1, (0, 0, 255), 4)
# cv2.circle(frame, (resize_b[7], resize_b[8]), 1, (0, 255, 255), 4)
# cv2.circle(frame, (resize_b[9], resize_b[10]), 1, (255, 0, 255), 4)
# cv2.circle(frame, (resize_b[11], resize_b[12]), 1, (0, 255, 0), 4)
# cv2.circle(frame, (resize_b[13], resize_b[14]), 1, (255, 0, 0), 4)
area = (resize_b[2] - resize_b[0]) * (resize_b[3] - resize_b[1])
if area > max_area:
max_area = area
max_index = i
i += 1
face_bboxes.append([resize_b[0], resize_b[1], resize_b[2], resize_b[3]])
face_landmarks.append([(resize_b[5], resize_b[6]),
(resize_b[7], resize_b[8]),
(resize_b[9], resize_b[10]),
(resize_b[11], resize_b[12]),
(resize_b[13], resize_b[14])])
# import time
# cv2.imwrite('results/0.jpg', frame)
return face_bboxes, face_landmarks, max_index
if __name__ == '__main__':
det_face_model_path = r'/home/jwq/PycharmProjects/situ/src/face_det/Pytorch_Retinaface/weights/mobilenet_0.25.mnn'
image_path = r'input/3.jpg'
image_save_path = r'results/3.jpg'
thr = 0.5
face_detector = Face_Detector(det_face_model_path)
image = cv2.imread(image_path)
face_detector.detect(image, thr)
# image_ploted = face_detector.plot(image, face_bboxes)
# cv2.imwrite(image_save_path, image_ploted)
Styling with Markdown is supported
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!