mirror of
https://github.com/iperov/DeepFaceLab.git
synced 2024-12-25 07:21:13 -08:00
0c2e1c3944
Maximum resolution is increased to 640. ‘hd’ archi is removed. ‘hd’ was experimental archi created to remove subpixel shake, but ‘lr_dropout’ and ‘disable random warping’ do that better. ‘uhd’ is renamed to ‘-u’ dfuhd and liaeuhd will be automatically renamed to df-u and liae-u in existing models. Added new experimental archi (key -d) which doubles the resolution using the same computation cost. It is mean same configs will be x2 faster, or for example you can set 448 resolution and it will train as 224. Strongly recommended not to train from scratch and use pretrained models. New archi naming: 'df' keeps more identity-preserved face. 'liae' can fix overly different face shapes. '-u' increased likeness of the face. '-d' (experimental) doubling the resolution using the same computation cost Examples: df, liae, df-d, df-ud, liae-ud, ... Improved GAN training (GAN_power option). It was used for dst model, but actually we don’t need it for dst. Instead, a second src GAN model with x2 smaller patch size was added, so the overall quality for hi-res models should be higher. Added option ‘Uniform yaw distribution of samples (y/n)’: Helps to fix blurry side faces due to small amount of them in the faceset. Quick96: Now based on df-ud archi and 20% faster. XSeg trainer: Improved sample generator. Now it randomly adds the background from other samples. Result is reduced chance of random mask noise on the area outside the face. Now you can specify ‘batch_size’ in range 2-16. Reduced size of samples with applied XSeg mask. Thus size of packed samples with applied xseg mask is also reduced.
304 lines
8.7 KiB
Python
304 lines
8.7 KiB
Python
from .MPSharedList import MPSharedList
|
|
import multiprocessing
|
|
import threading
|
|
import time
|
|
|
|
import numpy as np
|
|
|
|
|
|
class IndexHost():
|
|
"""
|
|
Provides random shuffled indexes for multiprocesses
|
|
"""
|
|
def __init__(self, indexes_count, rnd_seed=None):
|
|
self.sq = multiprocessing.Queue()
|
|
self.cqs = []
|
|
self.clis = []
|
|
self.thread = threading.Thread(target=self.host_thread, args=(indexes_count,rnd_seed) )
|
|
self.thread.daemon = True
|
|
self.thread.start()
|
|
|
|
def host_thread(self, indexes_count, rnd_seed):
|
|
rnd_state = np.random.RandomState(rnd_seed) if rnd_seed is not None else np.random
|
|
|
|
idxs = [*range(indexes_count)]
|
|
shuffle_idxs = []
|
|
sq = self.sq
|
|
|
|
while True:
|
|
while not sq.empty():
|
|
obj = sq.get()
|
|
cq_id, count = obj[0], obj[1]
|
|
|
|
result = []
|
|
for i in range(count):
|
|
if len(shuffle_idxs) == 0:
|
|
shuffle_idxs = idxs.copy()
|
|
rnd_state.shuffle(shuffle_idxs)
|
|
result.append(shuffle_idxs.pop())
|
|
self.cqs[cq_id].put (result)
|
|
|
|
time.sleep(0.001)
|
|
|
|
def create_cli(self):
|
|
cq = multiprocessing.Queue()
|
|
self.cqs.append ( cq )
|
|
cq_id = len(self.cqs)-1
|
|
return IndexHost.Cli(self.sq, cq, cq_id)
|
|
|
|
# disable pickling
|
|
def __getstate__(self):
|
|
return dict()
|
|
def __setstate__(self, d):
|
|
self.__dict__.update(d)
|
|
|
|
class Cli():
|
|
def __init__(self, sq, cq, cq_id):
|
|
self.sq = sq
|
|
self.cq = cq
|
|
self.cq_id = cq_id
|
|
|
|
def multi_get(self, count):
|
|
self.sq.put ( (self.cq_id,count) )
|
|
|
|
while True:
|
|
if not self.cq.empty():
|
|
return self.cq.get()
|
|
time.sleep(0.001)
|
|
|
|
class Index2DHost():
|
|
"""
|
|
Provides random shuffled indexes for multiprocesses
|
|
"""
|
|
def __init__(self, indexes2D):
|
|
self.sq = multiprocessing.Queue()
|
|
self.cqs = []
|
|
self.clis = []
|
|
self.thread = threading.Thread(target=self.host_thread, args=(indexes2D,) )
|
|
self.thread.daemon = True
|
|
self.thread.start()
|
|
|
|
def host_thread(self, indexes2D):
|
|
indexes2D_len = len(indexes2D)
|
|
|
|
idxs = [*range(indexes2D_len)]
|
|
idxs_2D = [None]*indexes2D_len
|
|
shuffle_idxs = []
|
|
shuffle_idxs_2D = [None]*indexes2D_len
|
|
for i in range(indexes2D_len):
|
|
idxs_2D[i] = [*range(len(indexes2D[i]))]
|
|
shuffle_idxs_2D[i] = []
|
|
|
|
#print(idxs)
|
|
#print(idxs_2D)
|
|
sq = self.sq
|
|
|
|
while True:
|
|
while not sq.empty():
|
|
obj = sq.get()
|
|
cq_id, count = obj[0], obj[1]
|
|
|
|
result = []
|
|
for i in range(count):
|
|
if len(shuffle_idxs) == 0:
|
|
shuffle_idxs = idxs.copy()
|
|
np.random.shuffle(shuffle_idxs)
|
|
|
|
idx_1D = shuffle_idxs.pop()
|
|
|
|
#print(f'idx_1D = {idx_1D}, len(shuffle_idxs_2D[idx_1D])= {len(shuffle_idxs_2D[idx_1D])}')
|
|
|
|
if len(shuffle_idxs_2D[idx_1D]) == 0:
|
|
shuffle_idxs_2D[idx_1D] = idxs_2D[idx_1D].copy()
|
|
#print(f'new shuffle_idxs_2d for {idx_1D} = { shuffle_idxs_2D[idx_1D] }')
|
|
|
|
#print(f'len(shuffle_idxs_2D[idx_1D])= {len(shuffle_idxs_2D[idx_1D])}')
|
|
|
|
np.random.shuffle( shuffle_idxs_2D[idx_1D] )
|
|
|
|
idx_2D = shuffle_idxs_2D[idx_1D].pop()
|
|
|
|
#print(f'len(shuffle_idxs_2D[idx_1D])= {len(shuffle_idxs_2D[idx_1D])}')
|
|
|
|
#print(f'idx_2D = {idx_2D}')
|
|
|
|
|
|
result.append( indexes2D[idx_1D][idx_2D])
|
|
|
|
self.cqs[cq_id].put (result)
|
|
|
|
time.sleep(0.001)
|
|
|
|
def create_cli(self):
|
|
cq = multiprocessing.Queue()
|
|
self.cqs.append ( cq )
|
|
cq_id = len(self.cqs)-1
|
|
return Index2DHost.Cli(self.sq, cq, cq_id)
|
|
|
|
# disable pickling
|
|
def __getstate__(self):
|
|
return dict()
|
|
def __setstate__(self, d):
|
|
self.__dict__.update(d)
|
|
|
|
class Cli():
|
|
def __init__(self, sq, cq, cq_id):
|
|
self.sq = sq
|
|
self.cq = cq
|
|
self.cq_id = cq_id
|
|
|
|
def multi_get(self, count):
|
|
self.sq.put ( (self.cq_id,count) )
|
|
|
|
while True:
|
|
if not self.cq.empty():
|
|
return self.cq.get()
|
|
time.sleep(0.001)
|
|
|
|
class ListHost():
|
|
def __init__(self, list_):
|
|
self.sq = multiprocessing.Queue()
|
|
self.cqs = []
|
|
self.clis = []
|
|
self.m_list = list_
|
|
self.thread = threading.Thread(target=self.host_thread)
|
|
self.thread.daemon = True
|
|
self.thread.start()
|
|
|
|
def host_thread(self):
|
|
sq = self.sq
|
|
while True:
|
|
while not sq.empty():
|
|
obj = sq.get()
|
|
cq_id, cmd = obj[0], obj[1]
|
|
|
|
if cmd == 0:
|
|
self.cqs[cq_id].put ( len(self.m_list) )
|
|
elif cmd == 1:
|
|
idx = obj[2]
|
|
item = self.m_list[idx ]
|
|
self.cqs[cq_id].put ( item )
|
|
elif cmd == 2:
|
|
result = []
|
|
for item in obj[2]:
|
|
result.append ( self.m_list[item] )
|
|
self.cqs[cq_id].put ( result )
|
|
elif cmd == 3:
|
|
self.m_list.insert(obj[2], obj[3])
|
|
elif cmd == 4:
|
|
self.m_list.append(obj[2])
|
|
elif cmd == 5:
|
|
self.m_list.extend(obj[2])
|
|
|
|
time.sleep(0.005)
|
|
|
|
def create_cli(self):
|
|
cq = multiprocessing.Queue()
|
|
self.cqs.append ( cq )
|
|
cq_id = len(self.cqs)-1
|
|
return ListHost.Cli(self.sq, cq, cq_id)
|
|
|
|
def get_list(self):
|
|
return self.list_
|
|
|
|
# disable pickling
|
|
def __getstate__(self):
|
|
return dict()
|
|
def __setstate__(self, d):
|
|
self.__dict__.update(d)
|
|
|
|
class Cli():
|
|
def __init__(self, sq, cq, cq_id):
|
|
self.sq = sq
|
|
self.cq = cq
|
|
self.cq_id = cq_id
|
|
|
|
def __len__(self):
|
|
self.sq.put ( (self.cq_id,0) )
|
|
|
|
while True:
|
|
if not self.cq.empty():
|
|
return self.cq.get()
|
|
time.sleep(0.001)
|
|
|
|
def __getitem__(self, key):
|
|
self.sq.put ( (self.cq_id,1,key) )
|
|
|
|
while True:
|
|
if not self.cq.empty():
|
|
return self.cq.get()
|
|
time.sleep(0.001)
|
|
|
|
def multi_get(self, keys):
|
|
self.sq.put ( (self.cq_id,2,keys) )
|
|
|
|
while True:
|
|
if not self.cq.empty():
|
|
return self.cq.get()
|
|
time.sleep(0.001)
|
|
|
|
def insert(self, index, item):
|
|
self.sq.put ( (self.cq_id,3,index,item) )
|
|
|
|
def append(self, item):
|
|
self.sq.put ( (self.cq_id,4,item) )
|
|
|
|
def extend(self, items):
|
|
self.sq.put ( (self.cq_id,5,items) )
|
|
|
|
|
|
|
|
class DictHost():
|
|
def __init__(self, d, num_users):
|
|
self.sqs = [ multiprocessing.Queue() for _ in range(num_users) ]
|
|
self.cqs = [ multiprocessing.Queue() for _ in range(num_users) ]
|
|
|
|
self.thread = threading.Thread(target=self.host_thread, args=(d,) )
|
|
self.thread.daemon = True
|
|
self.thread.start()
|
|
|
|
self.clis = [ DictHostCli(sq,cq) for sq, cq in zip(self.sqs, self.cqs) ]
|
|
|
|
def host_thread(self, d):
|
|
while True:
|
|
for sq, cq in zip(self.sqs, self.cqs):
|
|
if not sq.empty():
|
|
obj = sq.get()
|
|
cmd = obj[0]
|
|
if cmd == 0:
|
|
cq.put (d[ obj[1] ])
|
|
elif cmd == 1:
|
|
cq.put ( list(d.keys()) )
|
|
|
|
time.sleep(0.005)
|
|
|
|
|
|
def get_cli(self, n_user):
|
|
return self.clis[n_user]
|
|
|
|
# disable pickling
|
|
def __getstate__(self):
|
|
return dict()
|
|
def __setstate__(self, d):
|
|
self.__dict__.update(d)
|
|
|
|
class DictHostCli():
|
|
def __init__(self, sq, cq):
|
|
self.sq = sq
|
|
self.cq = cq
|
|
|
|
def __getitem__(self, key):
|
|
self.sq.put ( (0,key) )
|
|
|
|
while True:
|
|
if not self.cq.empty():
|
|
return self.cq.get()
|
|
time.sleep(0.001)
|
|
|
|
def keys(self):
|
|
self.sq.put ( (1,) )
|
|
while True:
|
|
if not self.cq.empty():
|
|
return self.cq.get()
|
|
time.sleep(0.001)
|