github/DeepFaceLab
1
0
Fork 0
mirror of https://github.com/iperov/DeepFaceLab.git synced 2025-07-07 21:42:08 -07:00
Code Issues Projects Releases Packages Wiki Activity Actions

upd dev_poseestimator

Browse source
This commit is contained in:
iperov 2019-04-24 11:33:11 +04:00
parent 947feac047
commit 60ed56a801
1 changed files with 10 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

12
facelib/PoseEstimator.py
View file

@ -21,7 +21,7 @@ class PoseEstimator(object):
exec( nnlib.import_all(), locals(), globals() ) exec( nnlib.import_all(), locals(), globals() )
self.angles = [90, 45, 30, 10, 2] self.angles = [90, 45, 30, 10, 2]
self.alpha_cat_losses = [0.07,0.05,0.03,0.01,0.01] self.alpha_cat_losses = [7,5,3,1,1]
self.class_nums = [ angle+1 for angle in self.angles ] self.class_nums = [ angle+1 for angle in self.angles ]
self.model = PoseEstimator.BuildModel(resolution, class_nums=self.class_nums) self.model = PoseEstimator.BuildModel(resolution, class_nums=self.class_nums)
@ -36,6 +36,12 @@ class PoseEstimator(object):
if load_weights: if load_weights:
self.model.load_weights (str(self.weights_path)) self.model.load_weights (str(self.weights_path))
else:
conv_weights_list = []
for layer in self.model.layers:
if type(layer) == keras.layers.Conv2D:
conv_weights_list += [layer.weights[0]] #Conv2D kernel_weights
CAInitializerMP ( conv_weights_list )
inp_t, = self.model.inputs inp_t, = self.model.inputs
bins_t = self.model.outputs bins_t = self.model.outputs
@ -61,7 +67,7 @@ class PoseEstimator(object):
idx_tensor = K.constant( np.array([idx for idx in range(self.class_nums[0])], dtype=K.floatx() ) ) idx_tensor = K.constant( np.array([idx for idx in range(self.class_nums[0])], dtype=K.floatx() ) )
pitch_t, yaw_t, roll_t = K.sum ( bins_t[0] * idx_tensor, 1), K.sum ( bins_t[1] * idx_tensor, 1), K.sum ( bins_t[2] * idx_tensor, 1) pitch_t, yaw_t, roll_t = K.sum ( bins_t[0] * idx_tensor, 1), K.sum ( bins_t[1] * idx_tensor, 1), K.sum ( bins_t[2] * idx_tensor, 1)
reg_alpha = 0.02 reg_alpha = 2
reg_pitch_loss = reg_alpha * K.mean(K.square( inp_pitch_t - pitch_t), -1) reg_pitch_loss = reg_alpha * K.mean(K.square( inp_pitch_t - pitch_t), -1)
reg_yaw_loss = reg_alpha * K.mean(K.square( inp_yaw_t - yaw_t), -1) reg_yaw_loss = reg_alpha * K.mean(K.square( inp_yaw_t - yaw_t), -1)
reg_roll_loss = reg_alpha * K.mean(K.square( inp_roll_t - roll_t), -1) reg_roll_loss = reg_alpha * K.mean(K.square( inp_roll_t - roll_t), -1)
@ -70,7 +76,7 @@ class PoseEstimator(object):
yaw_loss = reg_yaw_loss + sum(loss_yaw) yaw_loss = reg_yaw_loss + sum(loss_yaw)
roll_loss = reg_roll_loss + sum(loss_roll) roll_loss = reg_roll_loss + sum(loss_roll)
opt = Adam(lr=0.001, tf_cpu_mode=0) opt = Adam(lr=0.000001)
if training: if training:
self.train = K.function ([inp_t, inp_pitch_t, inp_yaw_t, inp_roll_t] + inp_bins_t, self.train = K.function ([inp_t, inp_pitch_t, inp_yaw_t, inp_roll_t] + inp_bins_t,