github/DeepFaceLab
1
0
Fork 0
mirror of https://github.com/iperov/DeepFaceLab.git synced 2025-07-07 05:22:06 -07:00
Code Issues Projects Releases Packages Wiki Activity Actions

SAE: added test option: 'Apply random color transfer to src faceset'

Browse source
This commit is contained in:
iperov 2019-05-06 11:34:56 +04:00
parent bde700243c
commit a805f81142
9 changed files with 152 additions and 129 deletions
Download patch file Download diff file Expand all files Collapse all files

14
converters/ConverterMasked.py
View file

@ -1,15 +1,18 @@
import time
import traceback
from .Converter import Converter
from facelib import LandmarksProcessor
from facelib import FaceType
from facelib import FANSegmentator
import cv2
import numpy as np
import imagelib
from facelib import FaceType, FANSegmentator, LandmarksProcessor
from interact import interact as io
from joblib import SubprocessFunctionCaller
from utils.pickle_utils import AntiPickler
import time
from .Converter import Converter
'''
default_mode = {1:'overlay',
2:'hist-match',
@ -430,4 +433,3 @@ class ConverterMasked(Converter):
debugs += [out_img.copy()]
return debugs if debug else out_img

2
models/Model_DEV_FANSEG/Model.py
View file

@ -48,7 +48,7 @@ class Model(ModelBase):
SampleGeneratorFace(self.training_data_src_path, debug=self.is_debug(), batch_size=self.batch_size,
sample_process_options=SampleProcessor.Options(random_flip=True),
output_sample_types=[ { 'types': (t.IMG_WARPED_TRANSFORMED, face_type, t.MODE_BGR_SHUFFLE), 'resolution' : self.resolution, 'motion_blur':(25, 1) },
{ 'types': (t.IMG_WARPED_TRANSFORMED, face_type, t.MODE_M, t.FACE_MASK_FULL), 'resolution': self.resolution },
{ 'types': (t.IMG_WARPED_TRANSFORMED, face_type, t.MODE_M), 'resolution': self.resolution },
]),
SampleGeneratorFace(self.training_data_dst_path, debug=self.is_debug(), batch_size=self.batch_size,

2
models/Model_DF/Model.py
View file

@ -47,7 +47,7 @@ class Model(ModelBase):
t = SampleProcessor.Types
output_sample_types=[ { 'types': (t.IMG_WARPED_TRANSFORMED, t.FACE_TYPE_FULL, t.MODE_BGR), 'resolution':128},
{ 'types': (t.IMG_TRANSFORMED, t.FACE_TYPE_FULL, t.MODE_BGR), 'resolution':128},
{ 'types': (t.IMG_TRANSFORMED, t.FACE_TYPE_FULL, t.MODE_M, t.FACE_MASK_FULL), 'resolution':128} ]
{ 'types': (t.IMG_TRANSFORMED, t.FACE_TYPE_FULL, t.MODE_M), 'resolution':128} ]
self.set_training_data_generators ([
SampleGeneratorFace(self.training_data_src_path, sort_by_yaw_target_samples_path=self.training_data_dst_path if self.sort_by_yaw else None,

2
models/Model_H128/Model.py
View file

@ -57,7 +57,7 @@ class Model(ModelBase):
t = SampleProcessor.Types
output_sample_types=[ { 'types': (t.IMG_WARPED_TRANSFORMED, t.FACE_TYPE_HALF, t.MODE_BGR), 'resolution':128},
{ 'types': (t.IMG_TRANSFORMED, t.FACE_TYPE_HALF, t.MODE_BGR), 'resolution':128},
{ 'types': (t.IMG_TRANSFORMED, t.FACE_TYPE_HALF, t.MODE_M, t.FACE_MASK_FULL), 'resolution':128} ]
{ 'types': (t.IMG_TRANSFORMED, t.FACE_TYPE_HALF, t.MODE_M), 'resolution':128} ]
self.set_training_data_generators ([
SampleGeneratorFace(self.training_data_src_path, sort_by_yaw_target_samples_path=self.training_data_dst_path if self.sort_by_yaw else None,

2
models/Model_H64/Model.py
View file

@ -58,7 +58,7 @@ class Model(ModelBase):
t = SampleProcessor.Types
output_sample_types=[ { 'types': (t.IMG_WARPED_TRANSFORMED, t.FACE_TYPE_HALF, t.MODE_BGR), 'resolution':64},
{ 'types': (t.IMG_TRANSFORMED, t.FACE_TYPE_HALF, t.MODE_BGR), 'resolution':64},
{ 'types': (t.IMG_TRANSFORMED, t.FACE_TYPE_HALF, t.MODE_M, t.FACE_MASK_FULL), 'resolution':64} ]
{ 'types': (t.IMG_TRANSFORMED, t.FACE_TYPE_HALF, t.MODE_M), 'resolution':64} ]
self.set_training_data_generators ([
SampleGeneratorFace(self.training_data_src_path, sort_by_yaw_target_samples_path=self.training_data_dst_path if self.sort_by_yaw else None,

2
models/Model_LIAEF128/Model.py
View file

@ -52,7 +52,7 @@ class Model(ModelBase):
t = SampleProcessor.Types
output_sample_types=[ { 'types': (t.IMG_WARPED_TRANSFORMED, t.FACE_TYPE_FULL, t.MODE_BGR), 'resolution':128},
{ 'types': (t.IMG_TRANSFORMED, t.FACE_TYPE_FULL, t.MODE_BGR), 'resolution':128},
{ 'types': (t.IMG_TRANSFORMED, t.FACE_TYPE_FULL, t.MODE_M, t.FACE_MASK_FULL), 'resolution':128} ]
{ 'types': (t.IMG_TRANSFORMED, t.FACE_TYPE_FULL, t.MODE_M), 'resolution':128} ]
self.set_training_data_generators ([
SampleGeneratorFace(self.training_data_src_path, sort_by_yaw_target_samples_path=self.training_data_dst_path if self.sort_by_yaw else None,

22
models/Model_SAE/Model.py
View file

@ -87,10 +87,15 @@ class SAEModel(ModelBase):
default_bg_style_power = default_bg_style_power if is_first_run else self.options.get('bg_style_power', default_bg_style_power)
self.options['bg_style_power'] = np.clip ( io.input_number("Background style power ( 0.0 .. 100.0 ?:help skip:%.2f) : " % (default_bg_style_power), default_bg_style_power,
help_message="Learn to transfer image around face. This can make face more like dst. Enabling this option increases the chance of model collapse."), 0.0, 100.0 )
default_apply_random_ct = False if is_first_run else self.options.get('apply_random_ct', False)
self.options['apply_random_ct'] = io.input_bool ("Apply random color transfer to src faceset? (y/n, ?:help skip:%s) : " % (yn_str[default_apply_random_ct]), default_apply_random_ct, help_message="Increase variativity of src samples by apply RCT color transfer from random dst samples.")
else:
self.options['pixel_loss'] = self.options.get('pixel_loss', False)
self.options['face_style_power'] = self.options.get('face_style_power', default_face_style_power)
self.options['bg_style_power'] = self.options.get('bg_style_power', default_bg_style_power)
self.options['apply_random_ct'] = self.options.get('apply_random_ct', False)
if is_first_run:
self.options['pretrain'] = io.input_bool ("Pretrain the model? (y/n, ?:help skip:n) : ", False, help_message="Pretrain the model with large amount of various faces. This technique may help to train the fake with overly different face shapes and light conditions of src/dst data. Face will be look more like a morphed. To reduce the morph effect, some model files will be initialized but not be updated after pretrain: LIAE: inter_AB.h5 DF: encoder.h5. The longer you pretrain the model the more morphed face will look. After that, save and run the training again.")
@ -117,6 +122,7 @@ class SAEModel(ModelBase):
self.ms_count = ms_count = 3 if (self.options['multiscale_decoder']) else 1
apply_random_ct = self.options.get('apply_random_ct', False)
masked_training = True
warped_src = Input(bgr_shape)
@ -133,7 +139,7 @@ class SAEModel(ModelBase):
target_dstm_ar = [ Input ( ( mask_shape[0] // (2**i) ,)*2 + (mask_shape[-1],) ) for i in range(ms_count-1, -1, -1)]
common_flow_kwargs = { 'padding': 'zero',
'norm': 'norm',
'norm': '',
'act':'' }
models_list = []
weights_to_load = []
@ -345,10 +351,6 @@ class SAEModel(ModelBase):
t_mode_bgr = t.MODE_BGR if not self.pretrain else t.MODE_BGR_SHUFFLE
output_sample_types = [ {'types' : (t.IMG_WARPED_TRANSFORMED, face_type, t_mode_bgr), 'resolution':resolution} ]
output_sample_types += [ {'types' : (t.IMG_TRANSFORMED, face_type, t_mode_bgr), 'resolution': resolution // (2**i) } for i in range(ms_count)]
output_sample_types += [ {'types' : (t.IMG_TRANSFORMED, face_type, t.MODE_M, t.FACE_MASK_FULL), 'resolution': resolution // (2**i) } for i in range(ms_count)]
training_data_src_path = self.training_data_src_path
training_data_dst_path = self.training_data_dst_path
sort_by_yaw = self.sort_by_yaw
@ -360,13 +362,19 @@ class SAEModel(ModelBase):
self.set_training_data_generators ([
SampleGeneratorFace(training_data_src_path, sort_by_yaw_target_samples_path=training_data_dst_path if sort_by_yaw else None,
random_ct_samples_path=training_data_dst_path if apply_random_ct else None,
debug=self.is_debug(), batch_size=self.batch_size,
sample_process_options=SampleProcessor.Options(random_flip=self.random_flip, scale_range=np.array([-0.05, 0.05])+self.src_scale_mod / 100.0 ),
output_sample_types=output_sample_types ),
output_sample_types = [ {'types' : (t.IMG_WARPED_TRANSFORMED, face_type, t_mode_bgr), 'resolution':resolution, 'apply_ct': apply_random_ct} ] + \
[ {'types' : (t.IMG_TRANSFORMED, face_type, t_mode_bgr), 'resolution': resolution // (2**i), 'apply_ct': apply_random_ct } for i in range(ms_count)] + \
[ {'types' : (t.IMG_TRANSFORMED, face_type, t.MODE_M), 'resolution': resolution // (2**i) } for i in range(ms_count)]
),
SampleGeneratorFace(training_data_dst_path, debug=self.is_debug(), batch_size=self.batch_size,
sample_process_options=SampleProcessor.Options(random_flip=self.random_flip, ),
output_sample_types=output_sample_types )
output_sample_types = [ {'types' : (t.IMG_WARPED_TRANSFORMED, face_type, t_mode_bgr), 'resolution':resolution} ] + \
[ {'types' : (t.IMG_TRANSFORMED, face_type, t_mode_bgr), 'resolution': resolution // (2**i)} for i in range(ms_count)] + \
[ {'types' : (t.IMG_TRANSFORMED, face_type, t.MODE_M), 'resolution': resolution // (2**i) } for i in range(ms_count)])
])
#override

30
samplelib/SampleGeneratorFace.py
View file

@ -1,11 +1,14 @@
import traceback
import numpy as np
import cv2
import multiprocessing
from utils import iter_utils
from facelib import LandmarksProcessor
import traceback
import cv2
import numpy as np
from facelib import LandmarksProcessor
from samplelib import (SampleGeneratorBase, SampleLoader, SampleProcessor,
SampleType)
from utils import iter_utils
from samplelib import SampleType, SampleProcessor, SampleLoader, SampleGeneratorBase
'''
arg
@ -15,7 +18,7 @@ output_sample_types = [
]
'''
class SampleGeneratorFace(SampleGeneratorBase):
def __init__ (self, samples_path, debug, batch_size, sort_by_yaw=False, sort_by_yaw_target_samples_path=None, sample_process_options=SampleProcessor.Options(), output_sample_types=[], add_sample_idx=False, generators_count=2, generators_random_seed=None, **kwargs):
def __init__ (self, samples_path, debug, batch_size, sort_by_yaw=False, sort_by_yaw_target_samples_path=None, random_ct_samples_path=None, sample_process_options=SampleProcessor.Options(), output_sample_types=[], add_sample_idx=False, generators_count=2, generators_random_seed=None, **kwargs):
super().__init__(samples_path, debug, batch_size)
self.sample_process_options = sample_process_options
self.output_sample_types = output_sample_types
@ -35,12 +38,14 @@ class SampleGeneratorFace(SampleGeneratorBase):
samples = SampleLoader.load (self.sample_type, self.samples_path, sort_by_yaw_target_samples_path)
ct_samples = SampleLoader.load (SampleType.FACE, random_ct_samples_path) if random_ct_samples_path is not None else None
if self.debug:
self.generators_count = 1
self.generators = [iter_utils.ThisThreadGenerator ( self.batch_func, (0, samples) )]
self.generators = [iter_utils.ThisThreadGenerator ( self.batch_func, (0, samples, ct_samples) )]
else:
self.generators_count = min ( generators_count, len(samples) )
self.generators = [iter_utils.SubprocessGenerator ( self.batch_func, (i, samples[i::self.generators_count] ) ) for i in range(self.generators_count) ]
self.generators = [iter_utils.SubprocessGenerator ( self.batch_func, (i, samples[i::self.generators_count], ct_samples ) ) for i in range(self.generators_count) ]
self.generator_counter = -1
@ -53,7 +58,7 @@ class SampleGeneratorFace(SampleGeneratorBase):
return next(generator)
def batch_func(self, param ):
generator_id, samples = param
generator_id, samples, ct_samples = param
if self.generators_random_seed is not None:
np.random.seed ( self.generators_random_seed[generator_id] )
@ -61,6 +66,8 @@ class SampleGeneratorFace(SampleGeneratorBase):
samples_len = len(samples)
samples_idxs = [*range(samples_len)]
ct_samples_len = len(ct_samples) if ct_samples is not None else 0
if len(samples_idxs) == 0:
raise ValueError('No training data provided.')
@ -106,7 +113,8 @@ class SampleGeneratorFace(SampleGeneratorBase):
if sample is not None:
try:
x = SampleProcessor.process (sample, self.sample_process_options, self.output_sample_types, self.debug)
x = SampleProcessor.process (sample, self.sample_process_options, self.output_sample_types, self.debug,
ct_sample=ct_samples[np.random.randint(ct_samples_len)] if ct_samples is not None else None )
except:
raise Exception ("Exception occured in sample %s. Error: %s" % (sample.filename, traceback.format_exc() ) )

53
samplelib/SampleProcessor.py
View file

@ -29,9 +29,6 @@ opts:
'FACE_TYPE_HEAD' #currently unused
'FACE_TYPE_AVATAR' #currently unused
'FACE_MASK_FULL'
'FACE_MASK_EYES' #currently unused
'MODE_BGR' #BGR
'MODE_G' #Grayscale
'MODE_GGG' #3xGrayscale
@ -42,6 +39,8 @@ opts:
'motion_blur' : (chance_int, range) - chance 0..100 to apply to face (not mask), and range [1..3] where 3 is highest power of motion blur
'apply_ct' : bool
"""
class SampleProcessor(object):
@ -65,11 +64,6 @@ class SampleProcessor(object):
FACE_TYPE_AVATAR = 13 #currently unused
FACE_TYPE_END = 20
FACE_MASK_BEGIN = 20
FACE_MASK_FULL = 20
FACE_MASK_EYES = 21 #currently unused
FACE_MASK_END = 30
MODE_BEGIN = 40
MODE_BGR = 40 #BGR
MODE_G = 41 #Grayscale
@ -89,10 +83,12 @@ class SampleProcessor(object):
self.ty_range = ty_range
@staticmethod
def process (sample, sample_process_options, output_sample_types, debug):
def process (sample, sample_process_options, output_sample_types, debug, ct_sample=None):
SPTF = SampleProcessor.Types
sample_bgr = sample.load_bgr()
ct_sample_bgr = None
ct_sample_mask = None
h,w,c = sample_bgr.shape
is_face_sample = sample.landmarks is not None
@ -121,6 +117,7 @@ class SampleProcessor(object):
normalize_std_dev = opts.get('normalize_std_dev', False)
normalize_vgg = opts.get('normalize_vgg', False)
motion_blur = opts.get('motion_blur', None)
apply_ct = opts.get('apply_ct', False)
img_type = SPTF.NONE
target_face_type = SPTF.NONE
@ -131,8 +128,6 @@ class SampleProcessor(object):
img_type = t
elif t >= SPTF.FACE_TYPE_BEGIN and t < SPTF.FACE_TYPE_END:
target_face_type = t
elif t >= SPTF.FACE_MASK_BEGIN and t < SPTF.FACE_MASK_END:
face_mask_type = t
elif t >= SPTF.MODE_BEGIN and t < SPTF.MODE_END:
mode_type = t
@ -163,10 +158,11 @@ class SampleProcessor(object):
if mode_type == SPTF.NONE:
raise ValueError ('expected MODE_ type')
img = cached_images.get(img_type, {}).get(face_mask_type, None)
img = cached_images.get(img_type, None)
if img is None:
img = sample_bgr
mask = None
cur_sample = sample
if is_face_sample:
@ -179,7 +175,6 @@ class SampleProcessor(object):
dim = mb_range[ np.random.randint(len(mb_range) ) ]
img = imagelib.LinearMotionBlur (img, dim, np.random.randint(180) )
if face_mask_type == SPTF.FACE_MASK_FULL:
mask = cur_sample.load_fanseg_mask() #using fanseg_mask if exist
if mask is None:
@ -188,19 +183,16 @@ class SampleProcessor(object):
if cur_sample.ie_polys is not None:
cur_sample.ie_polys.overlay_mask(mask)
img = np.concatenate( (img, mask ), -1 )
elif face_mask_type == SPTF.FACE_MASK_EYES:
mask = LandmarksProcessor.get_image_eye_mask (img.shape, cur_sample.landmarks)
mask = np.expand_dims (cv2.blur (mask, ( w // 32, w // 32 ) ), -1)
mask[mask > 0.0] = 1.0
img = np.concatenate( (img, mask ), -1 )
warp = (img_type==SPTF.IMG_WARPED or img_type==SPTF.IMG_WARPED_TRANSFORMED)
transform = (img_type==SPTF.IMG_WARPED_TRANSFORMED or img_type==SPTF.IMG_TRANSFORMED)
flip = img_type != SPTF.IMG_WARPED
is_border_replicate = face_mask_type == SPTF.NONE
img = cached_images[img_type][face_mask_type] = imagelib.warp_by_params (params, img, warp, transform, flip, is_border_replicate)
img = imagelib.warp_by_params (params, img, warp, transform, flip, True)
if mask is not None:
mask = imagelib.warp_by_params (params, mask, warp, transform, flip, False)[...,np.newaxis]
img = np.concatenate( (img, mask ), -1 )
cached_images[img_type] = img
if is_face_sample and target_face_type != SPTF.NONE:
ft = SPTF_FACETYPE_TO_FACETYPE[target_face_type]
@ -217,9 +209,24 @@ class SampleProcessor(object):
start_y = rnd_state.randint(sub_size+1)
img = img[start_y:start_y+sub_size,start_x:start_x+sub_size,:]
img = np.clip(img, 0, 1)
img_bgr = img[...,0:3]
img_mask = img[...,3:4]
if apply_ct:
if ct_sample_bgr is None:
ct_sample_bgr = ct_sample.load_bgr()
ct_sample_mask = LandmarksProcessor.get_image_hull_mask (ct_sample_bgr.shape, ct_sample.landmarks)
ct_sample_bgr_resized = cv2.resize( ct_sample_bgr, (resolution,resolution), cv2.INTER_LINEAR )
ct_sample_mask_resized = cv2.resize( ct_sample_mask, (resolution,resolution), cv2.INTER_LINEAR )[...,np.newaxis]
img_bgr = imagelib.reinhard_color_transfer ( np.clip( (img_bgr*255) .astype(np.uint8), 0, 255),
np.clip( (ct_sample_bgr_resized*255).astype(np.uint8), 0, 255),
source_mask=img_mask, target_mask=ct_sample_mask_resized)
img_bgr = np.clip( img_bgr.astype(np.float32) / 255.0, 0.0, 1.0)
if normalize_std_dev:
img_bgr = (img_bgr - img_bgr.mean( (0,1)) ) / img_bgr.std( (0,1) )
elif normalize_vgg:
@ -239,8 +246,6 @@ class SampleProcessor(object):
elif mode_type == SPTF.MODE_GGG:
img = np.concatenate ( ( np.repeat ( np.expand_dims(cv2.cvtColor(img_bgr, cv2.COLOR_BGR2GRAY),-1), (3,), -1), img_mask), -1)
elif mode_type == SPTF.MODE_M and is_face_sample:
if face_mask_type == SPTF.NONE:
raise ValueError ('no face_mask_type defined')
img = img_mask
if not debug: