github/DeepFaceLab
1
0
Fork 0
mirror of https://github.com/iperov/DeepFaceLab.git synced 2025-07-07 13:32:09 -07:00
Code Issues Projects Releases Packages Wiki Activity Actions

SAE: revert back CA weights option

Browse source
This commit is contained in:
iperov 2019-04-24 13:44:12 +04:00
parent 0cd8dd7296
commit 9535a657d2
6 changed files with 8 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

BIN
doc/manual_en_google_translated.docx
View file

Binary file not shown.

BIN
doc/manual_en_google_translated.pdf
View file

Binary file not shown.

BIN
doc/manual_ru.pdf
View file

Binary file not shown.

BIN
doc/manual_ru_source.docx
View file

Binary file not shown.

11
facelib/PoseEstimator.py
View file

@ -23,7 +23,6 @@ class PoseEstimator(object):
self.angles = [90, 45, 30, 10, 2]
self.alpha_cat_losses = [7,5,3,1,1]
self.class_nums = [ angle+1 for angle in self.angles ]
self.model = PoseEstimator.BuildModel(resolution, class_nums=self.class_nums)
@ -57,7 +56,7 @@ class PoseEstimator(object):
loss_pitch = []
loss_yaw = []
loss_roll = []
for i,class_num in enumerate(self.class_nums):
a = self.alpha_cat_losses[i]
loss_pitch += [ a*K.categorical_crossentropy( inp_bins_t[i*3+0], bins_t[i*3+0] ) ]
@ -65,7 +64,12 @@ class PoseEstimator(object):
loss_roll += [ a*K.categorical_crossentropy( inp_bins_t[i*3+2], bins_t[i*3+2] ) ]
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)
pitch_t, yaw_t, roll_t = nnlib.tf.reduce_sum ( bins_t[0] * idx_tensor, 1), nnlib.tf.reduce_sum ( bins_t[1] * idx_tensor, 1), nnlib.tf.reduce_sum ( bins_t[2] * idx_tensor, 1)
reg_alpha = 2
reg_pitch_loss = reg_alpha * K.mean(K.square( inp_pitch_t - pitch_t), -1)
@ -75,7 +79,6 @@ class PoseEstimator(object):
pitch_loss = reg_pitch_loss + sum(loss_pitch)
yaw_loss = reg_yaw_loss + sum(loss_yaw)
roll_loss = reg_roll_loss + sum(loss_roll)
opt = Adam(lr=0.000001)
if training:

2
models/Model_SAE/Model.py
View file

@ -67,7 +67,7 @@ class SAEModel(ModelBase):
self.options['d_ch_dims'] = np.clip ( io.input_int("Decoder dims per channel (10-85 ?:help skip:%d) : " % (default_d_ch_dims) , default_d_ch_dims, help_message="More decoder dims help to get better details, but require more VRAM. You can fine-tune model size to fit your GPU." ), 10, 85 )
#self.options['remove_gray_border'] = io.input_bool ("Remove gray border? (y/n, ?:help skip:n) : ", False, help_message="Removes gray border of predicted face, but requires more computing resources.")
self.options['multiscale_decoder'] = io.input_bool ("Use multiscale decoder? (y/n, ?:help skip:n) : ", False, help_message="Multiscale decoder helps to get better details.")
self.options['ca_weights'] = io.input_bool ("Use CA weights? (y/n, ?:help skip: %s ) : " % (yn_str[def_ca_weights]), def_ca_weights, help_message="Initialize network with 'Convolution Aware' weights. This may help to achieve a higher accuracy model, but consumes time at first run.")
self.options['ca_weights'] = io.input_bool ("Use CA weights? (y/n, ?:help skip: %s ) : " % (yn_str[def_ca_weights]), def_ca_weights, help_message="Initialize network with 'Convolution Aware' weights. This may help to achieve a higher accuracy model, but consumes a time at first run.")
else:
self.options['ae_dims'] = self.options.get('ae_dims', default_ae_dims)
self.options['e_ch_dims'] = self.options.get('e_ch_dims', default_e_ch_dims)