def predict(self, im, dets):
h, w, c = im.shape
dets = convert_to_square(dets, self.expand)
dets[:, 0:4] = np.round(dets[:, 0:4])
[dy, edy, dx, edx, y, ey, x, ex, tmpw, tmph] = pad(dets, w, h)
num_boxes = dets.shape[0]
cropped_ims = np.zeros((num_boxes, self.net_size, self.net_size, 3), dtype=np.float32)
for i in range(num_boxes):
tmp = np.zeros((tmph[i], tmpw[i], 3), dtype=np.uint8)
tmp[dy[i]:edy[i] + 1, dx[i]:edx[i] + 1, :] = im[y[i]:ey[i] + 1, x[i]:ex[i] + 1, :]
cropped_ims[i, :, :, :] = (cv2.resize(tmp, (self.net_size, self.net_size))-127.5) / 128
# print(cropped_ims.shape)
# cropped_ims.tofile('test_img.bin')
# cls_scores, reg, landmark, animoji = self.L_ONet.predict(cropped_ims)
cls_scores, reg,eye_eyebrow,mouth,multi = self.L_ONet.predict(cropped_ims)
# print(reg.shape)
# print(reg)
#prob belongs to face
cls_scores = cls_scores[:,1]
keep_inds = np.where(cls_scores > self.threshold)[0]
if len(keep_inds) > 0:
#pickout filtered box
boxes = dets[keep_inds]
boxes[:, 4] = cls_scores[keep_inds]
reg = reg[keep_inds]
# landmark = landmark[keep_inds]
# animoji = animoji[keep_inds]
eye_eyebrow = eye_eyebrow[keep_inds]
mouth = mouth[keep_inds]
multi = multi[keep_inds]
else:
# return None, None, None, None, None
return None, None, None,None,None,None
wh = boxes[:,2:4] - boxes[:,0:2] + 1
xy = boxes[:,0:2]
# landmark = landmark.reshape(-1,5,2)*wh[:,None,:] + xy[:,None,:]
# animoji = animoji.reshape(-1,70,2)*wh[:,None,:] + xy[:,None,:]
# eye = eye.reshape(-1,18,2)*wh[:,None,:] + xy[:,None,:]
eye_eyebrow = eye_eyebrow.reshape(-1,36,2)*wh[:,None,:] + xy[:,None,:]
mouth = mouth.reshape(-1,20,2)*wh[:,None,:] + xy[:,None,:]
multi = multi.reshape(-1,42,2)*wh[:,None,:] + xy[:,None,:]
boxes_c = calibrate_box(boxes, reg)
boxes = boxes[py_nms(boxes, 0.6, "Minimum")]
keep = py_nms(boxes_c, 0.6, "Minimum")
boxes_c = boxes_c[keep]
# landmark = landmark[keep]
# animoji = animoji[keep]
# eye = eye[keep]
eye_eyebrow = eye_eyebrow[keep]
mouth = mouth[keep]
multi = multi[keep]
# return boxes, boxes_c, landmark, animoji, reg
return boxes, boxes_c, eye_eyebrow,mouth,multi, reg
def predict(self, im):
current_scale = self.current_scale
im_resized = self.processed_image(im, current_scale)
current_height, current_width, _ = im_resized.shape
# fcn
all_boxes = list()
# print('pnet')
# isprint = True
while min(current_height, current_width) > self.net_size:
cls_cls_map, reg = self.PNet.predict(im_resized)
boxes = self.generate_bbox(cls_cls_map[:, :, 1], reg,
current_scale)
current_scale *= self.scale_factor
im_resized = self.processed_image(im, current_scale)
current_height, current_width, _ = im_resized.shape
# if isprint:
# print(cls_cls_map.shape)
# print(cls_cls_map)
# print(reg.shape)
# print(reg)
# print(im_resized.shape)
# print(im_resized)
# isprint = False
if boxes.size == 0:
continue
keep = py_nms(boxes[:, :5], 0.5, 'Union')
boxes = boxes[keep]
all_boxes.append(boxes)
if len(all_boxes) == 0:
return None, None, None
all_boxes = np.vstack(all_boxes)
# merge the detection from first stage
keep = py_nms(all_boxes[:, 0:5], 0.7, 'Union')
all_boxes = all_boxes[keep]
boxes = all_boxes[:, :5]
bbw = all_boxes[:, 2] - all_boxes[:, 0] + 1
bbh = all_boxes[:, 3] - all_boxes[:, 1] + 1
# refine the boxes
boxes_c = np.vstack([
all_boxes[:, 0] + all_boxes[:, 5] * bbw,
all_boxes[:, 1] + all_boxes[:, 6] * bbh,
all_boxes[:, 2] + all_boxes[:, 7] * bbw,
all_boxes[:, 3] + all_boxes[:, 8] * bbh, all_boxes[:, 4]
])
boxes_c = boxes_c.T
return boxes, boxes_c, None
def predict(self, im, dets):
h, w, c = im.shape
dets = convert_to_square(dets)
dets[:, 0:4] = np.round(dets[:, 0:4])
[dy, edy, dx, edx, y, ey, x, ex, tmpw, tmph] = pad(dets, w, h)
num_boxes = dets.shape[0]
cropped_ims = np.zeros((num_boxes, self.net_size, self.net_size, 3),
dtype=np.float32)
for i in range(num_boxes):
tmp = np.zeros((tmph[i], tmpw[i], 3), dtype=np.uint8)
tmp[dy[i]:edy[i] + 1, dx[i]:edx[i] + 1, :] = im[y[i]:ey[i] + 1,
x[i]:ex[i] + 1, :]
cropped_ims[i, :, :, :] = (
cv2.resize(tmp, (self.net_size, self.net_size)) - 127.5) / 128
cls_scores, reg, _, _ = self.RNet.predict(cropped_ims)
cls_scores = cls_scores[:, 1]
keep_inds = np.where(cls_scores > self.threshold)[0]
if len(keep_inds) > 0:
boxes = dets[keep_inds]
boxes[:, 4] = cls_scores[keep_inds]
reg = reg[keep_inds]
else:
return None, None, None
keep = py_nms(boxes, 0.6)
boxes = boxes[keep]
boxes_c = calibrate_box(boxes, reg[keep])
return boxes, boxes_c, None
def predict(self, im, dets):
h, w, c = im.shape
dets = convert_to_square(dets)
dets[:, 0:4] = np.round(dets[:, 0:4])
[dy, edy, dx, edx, y, ey, x, ex, tmpw, tmph] = pad(dets, w, h)
num_boxes = dets.shape[0]
cropped_ims = np.zeros((num_boxes, self.net_size, self.net_size, 3),
dtype=np.float32)
for i in range(num_boxes):
tmp = np.zeros((tmph[i], tmpw[i], 3), dtype=np.uint8)
tmp[dy[i]:edy[i] + 1, dx[i]:edx[i] + 1, :] = im[y[i]:ey[i] + 1,
x[i]:ex[i] + 1, :]
cropped_ims[i, :, :, :] = (
cv2.resize(tmp, (self.net_size, self.net_size)) - 127.5) / 128
# cls_scores, reg,landmark,_ = self.ONet.predict(cropped_ims)
cls_scores, reg, _ = self.ONet.predict(cropped_ims)
#prob belongs to face
cls_scores = cls_scores[:, 1]
keep_inds = np.where(cls_scores > self.threshold)[0]
if len(keep_inds) > 0:
#pickout filtered box
boxes = dets[keep_inds]
boxes[:, 4] = cls_scores[keep_inds]
reg = reg[keep_inds]
# landmark = landmark[keep_inds]
else:
return None, None, None, None
wh = boxes[:, 2:4] - boxes[:, 0:2] + 1
xy = boxes[:, 0:2]
# landmark = landmark.reshape(-1,5,2)*wh[:,None,:] + xy[:,None,:]
boxes_c = calibrate_box(boxes, reg)
boxes = boxes[py_nms(boxes, 0.6, "Minimum")]
keep = py_nms(boxes_c, 0.6, "Minimum")
boxes_c = boxes_c[keep]
# landmark = landmark[keep]
# return boxes, boxes_c, landmark, None
return boxes, boxes_c, None
def predict(self, im, dets):
threshold=0.6
net_size = 24
h, w, c = im.shape
dets = convert_to_square(dets)
dets[:, 0:4] = np.round(dets[:, 0:4])
[dy, edy, dx, edx, y, ey, x, ex, tmpw, tmph] = pad(dets, w, h)
num_boxes = dets.shape[0]
cropped_ims = np.zeros((num_boxes, net_size, net_size, 3), dtype=np.float32)
for i in range(num_boxes):
tmp = np.zeros((tmph[i], tmpw[i], 3), dtype=np.uint8)
tmp[dy[i]:edy[i] + 1, dx[i]:edx[i] + 1, :] = im[y[i]:ey[i] + 1, x[i]:ex[i] + 1, :]
cropped_ims[i, :, :, :] = (cv2.resize(tmp, (net_size, net_size))-127.5) / 128
cls_scores = []
reg = []
for img_ in cropped_ims:
cls_scores_i, reg_i,_ = self.forward(img_.transpose(2,0,1),None)
cls_scores.append(cls_scores_i.reshape(-1))
reg.append(reg_i.reshape(-1))
cls_scores = np.asarray(cls_scores)
reg = np.asarray(reg)
cls_scores = cls_scores[:,1]
keep_inds = np.where(cls_scores > threshold)[0]
if len(keep_inds) > 0:
boxes = dets[keep_inds]
boxes[:, 4] = cls_scores[keep_inds]
reg = reg[keep_inds]
else:
return None, None, None
keep = py_nms(boxes, 0.6)
boxes = boxes[keep]
boxes_c = calibrate_box(boxes, reg[keep])
return boxes, boxes_c, None
def predict(self, im):
min_face_size=25
stride=2
threshold=0.6
scale_factor = 0.79
# self.min_face_size = min_face_size
# self.stride = stride
# self.threshold = threshold
# self.scale_factor = scale_factor
# # h, w, c = im.shape
net_size = 12
current_scale = float(net_size) / min_face_size # find initial scale
# print("current_scale", net_size, self.min_face_size, current_scale)
# current_scale = self.current_scale
im_resized = self.processed_image(im, current_scale)
current_height, current_width, _ = im_resized.shape
# print('numpys')
# print(im_resized.shape)
# print(im_resized)
# isprint = True
# fcn
all_boxes = list()
while min(current_height, current_width) > net_size:
cls_cls_map, reg,_ = self.forward(im_resized.transpose(2,0,1),None)
boxes = self.generate_bbox(cls_cls_map[1, :, :], reg, current_scale)
current_scale *= scale_factor
im_resized = self.processed_image(im, current_scale)
current_height, current_width, _ = im_resized.shape
# if isprint:
# print(cls_cls_map.shape)
# print(cls_cls_map.transpose(1,2,0))
# print(reg.shape)
# print(reg.transpose(1,2,0))
# print(im_resized.shape)
# print(im_resized)
# isprint = False
if boxes.size == 0:
continue
keep = py_nms(boxes[:, :5], 0.5, 'Union')
boxes = boxes[keep]
all_boxes.append(boxes)
if len(all_boxes) == 0:
return None, None, None
all_boxes = np.vstack(all_boxes)
# merge the detection from first stage
keep = py_nms(all_boxes[:, 0:5], 0.7, 'Union')
all_boxes = all_boxes[keep]
boxes = all_boxes[:, :5]
bbw = all_boxes[:, 2] - all_boxes[:, 0] + 1
bbh = all_boxes[:, 3] - all_boxes[:, 1] + 1
# refine the boxes
boxes_c = np.vstack([all_boxes[:, 0] + all_boxes[:, 5] * bbw,
all_boxes[:, 1] + all_boxes[:, 6] * bbh,
all_boxes[:, 2] + all_boxes[:, 7] * bbw,
all_boxes[:, 3] + all_boxes[:, 8] * bbh,
all_boxes[:, 4]])
boxes_c = boxes_c.T
return boxes, boxes_c, None
def predict(self, im, dets_):
dets = dets_.copy()
net_size = 48
threshold = 0.7
expand = 0.3
h, w, c = im.shape
dets = convert_to_square(dets, expand)
dets[:, 0:4] = np.round(dets[:, 0:4])
[dy, edy, dx, edx, y, ey, x, ex, tmpw, tmph] = pad(dets, w, h)
num_boxes = dets.shape[0]
cropped_ims = np.zeros((num_boxes, net_size, net_size, 3), dtype=np.float32)
for i in range(num_boxes):
tmp = np.zeros((tmph[i], tmpw[i], 3), dtype=np.uint8)
tmp[dy[i]:edy[i] + 1, dx[i]:edx[i] + 1, :] = im[y[i]:ey[i] + 1, x[i]:ex[i] + 1, :]
cropped_ims[i, :, :, :] = (cv2.resize(tmp, (net_size, net_size))-127.5) / 128
cls_scores = []
reg = []
landmark = []
animoji = []
for img_ in cropped_ims:
cls_scores_i, reg_i,landmark_i, animoji_i = self.forward(img_.transpose(2,0,1),None)
cls_scores.append(cls_scores_i.reshape(-1))
reg.append(reg_i.reshape(-1))
landmark.append(landmark_i.reshape(-1))
animoji.append(animoji_i.reshape(-1))
cls_scores = np.asarray(cls_scores)
reg = np.asarray(reg)
landmark = np.asarray(landmark)
animoji = np.asarray(animoji)
# print(reg.shape)
# print(reg)
# f.close()
#prob belongs to face
cls_scores = cls_scores[:,1]
keep_inds = np.where(cls_scores > threshold)[0]
if len(keep_inds) > 0:
#pickout filtered box
boxes = dets[keep_inds]
boxes[:, 4] = cls_scores[keep_inds]
reg = reg[keep_inds]
landmark = landmark[keep_inds]
animoji = animoji[keep_inds]
else:
return None, None, None
wh = boxes[:,2:4] - boxes[:,0:2] + 1
xy = boxes[:,0:2]
landmark = landmark.reshape(-1,5,2)*wh[:,None,:] + xy[:,None,:]
animoji = animoji.reshape(-1,70,2)*wh[:,None,:] + xy[:,None,:]
boxes_c = calibrate_box(boxes, reg)
boxes = boxes[py_nms(boxes, 0.6, "Minimum")]
keep = py_nms(boxes_c, 0.6, "Minimum")
boxes_c = boxes_c[keep]
landmark = landmark[keep]
return boxes, boxes_c, landmark, animoji,reg
