mirror of
https://github.com/iperov/DeepFaceLab.git
synced 2025-07-07 05:22:06 -07:00
refactorings, improved sort by hist-dissim
This commit is contained in:
iperov
parent
4ff67ad26b
commit
9926dc626a
6 changed files with 128 additions and 127 deletions
2
main.py
2
main.py
|
|
@ -58,7 +58,7 @@ if __name__ == "__main__":
|
|||
|
||||
sort_parser = subparsers.add_parser( "sort", help="Sort faces in a directory.")
|
||||
sort_parser.add_argument('--input-dir', required=True, action=fixPathAction, dest="input_dir", help="Input directory. A directory containing the files you wish to process.")
|
||||
sort_parser.add_argument('--by', required=True, dest="sort_by_method", choices=("blur", "face", "face-dissim", "face-yaw", "hist", "hist-dissim", "hist-blur", "ssim", "brightness", "hue", "black", "origname"), help="Method of sorting. 'origname' sort by original filename to recover original sequence." )
|
||||
sort_parser.add_argument('--by', required=True, dest="sort_by_method", choices=("blur", "face", "face-dissim", "face-yaw", "hist", "hist-dissim", "hist-blur", "brightness", "hue", "black", "origname"), help="Method of sorting. 'origname' sort by original filename to recover original sequence." )
|
||||
sort_parser.set_defaults (func=process_sort)
|
||||
|
||||
def process_train(arguments):
|
||||
|
|
|
|||
|
|
@ -3,7 +3,7 @@ from pathlib import Path
|
|||
from utils import Path_utils
|
||||
import cv2
|
||||
from tqdm import tqdm
|
||||
from utils.AlignedPNG import AlignedPNG
|
||||
from utils.DFLPNG import DFLPNG
|
||||
from utils import image_utils
|
||||
import shutil
|
||||
import numpy as np
|
||||
|
|
@ -156,12 +156,7 @@ class ConvertSubprocessor(SubprocessorBase):
|
|||
image = (cv2.imread(str(filename_path)) / 255.0).astype(np.float32)
|
||||
|
||||
if self.converter.get_mode() == ConverterBase.MODE_IMAGE:
|
||||
image_landmarks = None
|
||||
a_png = AlignedPNG.load( str(filename_path) )
|
||||
if a_png is not None:
|
||||
d = a_png.getFaceswapDictData()
|
||||
if d is not None and 'landmarks' in d.keys():
|
||||
image_landmarks = np.array(d['landmarks'])
|
||||
image_landmarks = DFLPNG.load( str(filename_path), throw_on_no_embedded_data=True ).get_landmarks()
|
||||
|
||||
image = self.converter.convert_image(image, image_landmarks, self.debug)
|
||||
if self.debug:
|
||||
|
|
@ -258,20 +253,15 @@ def main (input_dir, output_dir, model_dir, model_name, aligned_dir=None, **in_o
|
|||
|
||||
aligned_path_image_paths = Path_utils.get_image_paths(aligned_path)
|
||||
for filename in tqdm(aligned_path_image_paths, desc= "Collecting alignments" ):
|
||||
a_png = AlignedPNG.load( str(filename) )
|
||||
if a_png is None:
|
||||
print ( "%s - no embedded data found." % (filename) )
|
||||
continue
|
||||
d = a_png.getFaceswapDictData()
|
||||
if d is None or d['source_filename'] is None or d['source_rect'] is None or d['source_landmarks'] is None:
|
||||
print ( "%s - no embedded data found." % (filename) )
|
||||
dflpng = DFLPNG.load( str(filename), print_on_no_embedded_data=True )
|
||||
if dflpng is None:
|
||||
continue
|
||||
|
||||
source_filename_stem = Path(d['source_filename']).stem
|
||||
source_filename_stem = Path( dflpng.get_source_filename() ).stem
|
||||
if source_filename_stem not in alignments.keys():
|
||||
alignments[ source_filename_stem ] = []
|
||||
|
||||
alignments[ source_filename_stem ].append ( np.array(d['source_landmarks']) )
|
||||
alignments[ source_filename_stem ].append (dflpng.get_source_landmarks())
|
||||
|
||||
|
||||
files_processed, faces_processed = ConvertSubprocessor (
|
||||
|
|
|
|||
|
|
@ -8,7 +8,7 @@ from pathlib import Path
|
|||
import numpy as np
|
||||
import cv2
|
||||
from utils import Path_utils
|
||||
from utils.AlignedPNG import AlignedPNG
|
||||
from utils.DFLPNG import DFLPNG
|
||||
from utils import image_utils
|
||||
from facelib import FaceType
|
||||
import facelib
|
||||
|
|
@ -313,20 +313,15 @@ class ExtractSubprocessor(SubprocessorBase):
|
|||
face_image_landmarks = facelib.LandmarksProcessor.transform_points (image_landmarks, image_to_face_mat)
|
||||
|
||||
cv2.imwrite(output_file, face_image)
|
||||
|
||||
a_png = AlignedPNG.load (output_file)
|
||||
|
||||
d = {
|
||||
'face_type': FaceType.toString(self.face_type),
|
||||
'landmarks': face_image_landmarks.tolist(),
|
||||
'yaw_value': facelib.LandmarksProcessor.calc_face_yaw (face_image_landmarks),
|
||||
'pitch_value': facelib.LandmarksProcessor.calc_face_pitch (face_image_landmarks),
|
||||
'source_filename': filename_path.name,
|
||||
'source_rect': rect,
|
||||
'source_landmarks': image_landmarks.tolist()
|
||||
}
|
||||
a_png.setFaceswapDictData (d)
|
||||
a_png.save(output_file)
|
||||
|
||||
DFLPNG.embed_data(output_file, face_type = FaceType.toString(self.face_type),
|
||||
landmarks = face_image_landmarks.tolist(),
|
||||
yaw_value = facelib.LandmarksProcessor.calc_face_yaw (face_image_landmarks),
|
||||
pitch_value = facelib.LandmarksProcessor.calc_face_pitch (face_image_landmarks),
|
||||
source_filename = filename_path.name,
|
||||
source_rect= rect,
|
||||
source_landmarks = image_landmarks.tolist()
|
||||
)
|
||||
|
||||
result.append (output_file)
|
||||
|
||||
|
|
|
|||
|
|
@ -8,7 +8,8 @@ from shutil import copyfile
|
|||
|
||||
from pathlib import Path
|
||||
from utils import Path_utils
|
||||
from utils.AlignedPNG import AlignedPNG
|
||||
from utils import image_utils
|
||||
from utils.DFLPNG import DFLPNG
|
||||
from facelib import LandmarksProcessor
|
||||
from utils.SubprocessorBase import SubprocessorBase
|
||||
import multiprocessing
|
||||
|
|
@ -86,22 +87,16 @@ class BlurEstimatorSubprocessor(SubprocessorBase):
|
|||
#override
|
||||
def onClientProcessData(self, data):
|
||||
filename_path = Path( data[0] )
|
||||
image = cv2.imread( str(filename_path) )
|
||||
face_mask = None
|
||||
|
||||
a_png = AlignedPNG.load( str(filename_path) )
|
||||
if a_png is not None:
|
||||
d = a_png.getFaceswapDictData()
|
||||
if (d is not None) and (d['landmarks'] is not None):
|
||||
face_mask = LandmarksProcessor.get_image_hull_mask (image, np.array(d['landmarks']))
|
||||
|
||||
if face_mask is not None:
|
||||
image = (image*face_mask).astype(np.uint8)
|
||||
|
||||
dflpng = DFLPNG.load( str(filename_path), print_on_no_embedded_data=True )
|
||||
if dflpng is not None:
|
||||
image = cv2.imread( str(filename_path) )
|
||||
image = ( image * \
|
||||
LandmarksProcessor.get_image_hull_mask (image, dflpng.get_landmarks()) \
|
||||
).astype(np.uint8)
|
||||
return [ str(filename_path), estimate_sharpness( image ) ]
|
||||
else:
|
||||
print ( "%s - no embedded data found." % (str(filename_path)) )
|
||||
return [ str(filename_path), 0 ]
|
||||
|
||||
return [ str(filename_path), estimate_sharpness( image ) ]
|
||||
|
||||
#override
|
||||
def onClientGetDataName (self, data):
|
||||
|
|
@ -164,18 +159,11 @@ def sort_by_face(input_path):
|
|||
print ("%s is not a png file required for sort_by_face" % (filepath.name) )
|
||||
continue
|
||||
|
||||
a_png = AlignedPNG.load (str(filepath))
|
||||
if a_png is None:
|
||||
print ("%s failed to load" % (filepath.name) )
|
||||
continue
|
||||
|
||||
d = a_png.getFaceswapDictData()
|
||||
|
||||
if d is None or d['landmarks'] is None:
|
||||
print ("%s - no embedded data found required for sort_by_face" % (filepath.name) )
|
||||
dflpng = DFLPNG.load (str(filepath), print_on_no_embedded_data=True)
|
||||
if dflpng is None:
|
||||
continue
|
||||
|
||||
img_list.append( [str(filepath), np.array(d['landmarks']) ] )
|
||||
img_list.append( [str(filepath), dflpng.get_landmarks()] )
|
||||
|
||||
|
||||
img_list_len = len(img_list)
|
||||
|
|
@ -207,18 +195,11 @@ def sort_by_face_dissim(input_path):
|
|||
print ("%s is not a png file required for sort_by_face_dissim" % (filepath.name) )
|
||||
continue
|
||||
|
||||
a_png = AlignedPNG.load (str(filepath))
|
||||
if a_png is None:
|
||||
print ("%s failed to load" % (filepath.name) )
|
||||
continue
|
||||
|
||||
d = a_png.getFaceswapDictData()
|
||||
|
||||
if d is None or d['landmarks'] is None:
|
||||
print ("%s - no embedded data found required for sort_by_face_dissim" % (filepath.name) )
|
||||
dflpng = DFLPNG.load (str(filepath), print_on_no_embedded_data=True)
|
||||
if dflpng is None:
|
||||
continue
|
||||
|
||||
img_list.append( [str(filepath), np.array(d['landmarks']), 0 ] )
|
||||
img_list.append( [str(filepath), dflpng.get_landmarks(), 0 ] )
|
||||
|
||||
img_list_len = len(img_list)
|
||||
for i in tqdm( range(0, img_list_len-1), desc="Sorting"):
|
||||
|
|
@ -247,18 +228,11 @@ def sort_by_face_yaw(input_path):
|
|||
print ("%s is not a png file required for sort_by_face_dissim" % (filepath.name) )
|
||||
continue
|
||||
|
||||
a_png = AlignedPNG.load (str(filepath))
|
||||
if a_png is None:
|
||||
print ("%s failed to load" % (filepath.name) )
|
||||
continue
|
||||
|
||||
d = a_png.getFaceswapDictData()
|
||||
|
||||
if d is None or d['yaw_value'] is None:
|
||||
print ("%s - no embedded data found required for sort_by_face_dissim" % (filepath.name) )
|
||||
dflpng = DFLPNG.load (str(filepath), print_on_no_embedded_data=True)
|
||||
if dflpng is None:
|
||||
continue
|
||||
|
||||
img_list.append( [str(filepath), np.array(d['yaw_value']) ] )
|
||||
img_list.append( [str(filepath), np.array( dflpng.get_yaw_value() ) ] )
|
||||
|
||||
print ("Sorting...")
|
||||
img_list = sorted(img_list, key=operator.itemgetter(1), reverse=True)
|
||||
|
|
@ -423,9 +397,7 @@ class HistDissimSubprocessor(SubprocessorBase):
|
|||
for j in range( 0, self.img_list_len):
|
||||
if i == j:
|
||||
continue
|
||||
score_total += cv2.compareHist(self.img_list[i][1], self.img_list[j][1], cv2.HISTCMP_BHATTACHARYYA) + \
|
||||
cv2.compareHist(self.img_list[i][2], self.img_list[j][2], cv2.HISTCMP_BHATTACHARYYA) + \
|
||||
cv2.compareHist(self.img_list[i][3], self.img_list[j][3], cv2.HISTCMP_BHATTACHARYYA)
|
||||
score_total += cv2.compareHist(self.img_list[i][1], self.img_list[j][1], cv2.HISTCMP_BHATTACHARYYA)
|
||||
|
||||
return score_total
|
||||
|
||||
|
|
@ -436,7 +408,7 @@ class HistDissimSubprocessor(SubprocessorBase):
|
|||
|
||||
#override
|
||||
def onHostResult (self, data, result):
|
||||
self.img_list[data[0]][4] = result
|
||||
self.img_list[data[0]][2] = result
|
||||
return 1
|
||||
|
||||
#override
|
||||
|
|
@ -451,17 +423,20 @@ def sort_by_hist_dissim(input_path):
|
|||
print ("Sorting by histogram dissimilarity...")
|
||||
|
||||
img_list = []
|
||||
for x in tqdm( Path_utils.get_image_paths(input_path), desc="Loading"):
|
||||
img = cv2.imread(x)
|
||||
img_list.append ([x, cv2.calcHist([img], [0], None, [256], [0, 256]),
|
||||
cv2.calcHist([img], [1], None, [256], [0, 256]),
|
||||
cv2.calcHist([img], [2], None, [256], [0, 256]), 0
|
||||
])
|
||||
for filename_path in tqdm( Path_utils.get_image_paths(input_path), desc="Loading"):
|
||||
image = cv2.imread(filename_path)
|
||||
|
||||
dflpng = DFLPNG.load( str(filename_path), print_on_no_embedded_data=True )
|
||||
if dflpng is not None:
|
||||
face_mask = LandmarksProcessor.get_image_hull_mask (image, dflpng.get_landmarks())
|
||||
image = (image*face_mask).astype(np.uint8)
|
||||
|
||||
img_list.append ([filename_path, cv2.calcHist([cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)], [0], None, [256], [0, 256]), 0 ])
|
||||
|
||||
img_list = HistDissimSubprocessor(img_list).process()
|
||||
|
||||
print ("Sorting...")
|
||||
img_list = sorted(img_list, key=operator.itemgetter(4), reverse=True)
|
||||
img_list = sorted(img_list, key=operator.itemgetter(2), reverse=True)
|
||||
|
||||
return img_list
|
||||
|
||||
|
|
@ -508,18 +483,11 @@ def sort_by_origname(input_path):
|
|||
print ("%s is not a png file required for sort_by_origname" % (filepath.name) )
|
||||
continue
|
||||
|
||||
a_png = AlignedPNG.load (str(filepath))
|
||||
if a_png is None:
|
||||
print ("%s failed to load" % (filepath.name) )
|
||||
continue
|
||||
|
||||
d = a_png.getFaceswapDictData()
|
||||
|
||||
if d is None or d['source_filename'] is None:
|
||||
print ("%s - no embedded data found required for sort_by_origname" % (filepath.name) )
|
||||
dflpng = DFLPNG.load (str(filepath), print_on_no_embedded_data=True)
|
||||
if dflpng is None:
|
||||
continue
|
||||
|
||||
img_list.append( [str(filepath), d['source_filename']] )
|
||||
img_list.append( [str(filepath), dflpng.get_source_filename()] )
|
||||
|
||||
print ("Sorting...")
|
||||
img_list = sorted(img_list, key=operator.itemgetter(1))
|
||||
|
|
@ -545,4 +513,4 @@ def main (input_path, sort_by_method):
|
|||
elif sort_by_method == 'black': img_list = sort_by_black (input_path)
|
||||
elif sort_by_method == 'origname': img_list = sort_by_origname (input_path)
|
||||
|
||||
final_rename (input_path, img_list)
|
||||
final_rename (input_path, img_list)
|
||||
|
|
|
|||
|
|
@ -4,7 +4,7 @@ from pathlib import Path
|
|||
from tqdm import tqdm
|
||||
import numpy as np
|
||||
import cv2
|
||||
from utils.AlignedPNG import AlignedPNG
|
||||
from utils.DFLPNG import DFLPNG
|
||||
from utils import iter_utils
|
||||
from utils import Path_utils
|
||||
from .BaseTypes import TrainingDataType
|
||||
|
|
@ -177,19 +177,14 @@ def X_LOAD ( RAWS ):
|
|||
print ("%s is not a png file required for training" % (s_filename_path.name) )
|
||||
continue
|
||||
|
||||
a_png = AlignedPNG.load ( str(s_filename_path) )
|
||||
if a_png is None:
|
||||
print ("%s failed to load" % (s_filename_path.name) )
|
||||
dflpng = DFLPNG.load ( str(s_filename_path), print_on_no_embedded_data=True )
|
||||
if dflpng is None:
|
||||
continue
|
||||
|
||||
d = a_png.getFaceswapDictData()
|
||||
if d is None or d['landmarks'] is None or d['yaw_value'] is None:
|
||||
print ("%s - no embedded faceswap info found required for training" % (s_filename_path.name) )
|
||||
continue
|
||||
|
||||
face_type = d['face_type'] if 'face_type' in d.keys() else 'full_face'
|
||||
face_type = FaceType.fromString (face_type)
|
||||
sample_list.append( s.copy_and_set(face_type=face_type, shape=a_png.get_shape(), landmarks=d['landmarks'], yaw=d['yaw_value']) )
|
||||
sample_list.append( s.copy_and_set(face_type=FaceType.fromString (dflpng.get_face_type()),
|
||||
shape=dflpng.get_shape(),
|
||||
landmarks=dflpng.get_landmarks(),
|
||||
yaw=dflpng.get_yaw_value()) )
|
||||
|
||||
return sample_list
|
||||
|
||||
|
|
|
|||
|
|
@ -4,6 +4,7 @@ import string
|
|||
import struct
|
||||
import zlib
|
||||
import pickle
|
||||
import numpy as np
|
||||
|
||||
class Chunk(object):
|
||||
def __init__(self, name=None, data=None):
|
||||
|
|
@ -184,7 +185,7 @@ class IEND(Chunk):
|
|||
def __str__(self):
|
||||
return "<Chunk:IEND>".format(**self.__dict__)
|
||||
|
||||
class FaceswapChunk(Chunk):
|
||||
class DFLChunk(Chunk):
|
||||
def __init__(self, dict_data=None):
|
||||
super().__init__("fcWp")
|
||||
self.dict_data = dict_data
|
||||
|
|
@ -207,26 +208,26 @@ class FaceswapChunk(Chunk):
|
|||
|
||||
chunk_map = {
|
||||
b"IHDR": IHDR,
|
||||
b"fcWp": FaceswapChunk,
|
||||
b"fcWp": DFLChunk,
|
||||
b"IEND": IEND
|
||||
}
|
||||
|
||||
class AlignedPNG(object):
|
||||
class DFLPNG(object):
|
||||
def __init__(self):
|
||||
self.data = b""
|
||||
self.length = 0
|
||||
self.chunks = []
|
||||
|
||||
self.fcwp_dict = None
|
||||
|
||||
@staticmethod
|
||||
def load(data):
|
||||
|
||||
def load_raw(filename):
|
||||
try:
|
||||
with open(data, "rb") as f:
|
||||
with open(filename, "rb") as f:
|
||||
data = f.read()
|
||||
except:
|
||||
raise FileNotFoundError(data)
|
||||
|
||||
inst = AlignedPNG()
|
||||
inst = DFLPNG()
|
||||
inst.data = data
|
||||
inst.length = len(data)
|
||||
|
||||
|
|
@ -242,14 +243,47 @@ class AlignedPNG(object):
|
|||
chunk = chunk_map.get(chunk_name, Chunk).load(data[chunk_start:chunk_end])
|
||||
inst.chunks.append(chunk)
|
||||
chunk_start = chunk_end
|
||||
|
||||
|
||||
return inst
|
||||
|
||||
@staticmethod
|
||||
def load(filename, print_on_no_embedded_data=False, throw_on_no_embedded_data=False):
|
||||
inst = DFLPNG.load_raw (filename)
|
||||
inst.fcwp_dict = inst.getDFLDictData()
|
||||
|
||||
def save(self, filename):
|
||||
if inst.fcwp_dict == None:
|
||||
if print_on_no_embedded_data:
|
||||
print ( "No DFL data found in %s" % (filename) )
|
||||
if throw_on_no_embedded_data:
|
||||
raise ValueError("No DFL data found in %s" % (filename) )
|
||||
return None
|
||||
|
||||
return inst
|
||||
|
||||
@staticmethod
|
||||
def embed_data(filename, face_type=None,
|
||||
landmarks=None,
|
||||
yaw_value=None,
|
||||
pitch_value=None,
|
||||
source_filename=None,
|
||||
source_rect=None,
|
||||
source_landmarks=None
|
||||
):
|
||||
|
||||
inst = DFLPNG.load_raw (filename)
|
||||
inst.setDFLDictData ({
|
||||
'face_type': face_type,
|
||||
'landmarks': landmarks,
|
||||
'yaw_value': yaw_value,
|
||||
'pitch_value': pitch_value,
|
||||
'source_filename': source_filename,
|
||||
'source_rect': source_rect,
|
||||
'source_landmarks': source_landmarks
|
||||
})
|
||||
|
||||
try:
|
||||
with open(filename, "wb") as f:
|
||||
f.write ( self.dump() )
|
||||
f.write ( inst.dump() )
|
||||
except:
|
||||
raise Exception( 'cannot save %s' % (filename) )
|
||||
|
||||
|
|
@ -274,23 +308,42 @@ class AlignedPNG(object):
|
|||
return chunk.height
|
||||
return 0
|
||||
|
||||
def getFaceswapDictData(self):
|
||||
def getDFLDictData(self):
|
||||
for chunk in self.chunks:
|
||||
if type(chunk) == FaceswapChunk:
|
||||
if type(chunk) == DFLChunk:
|
||||
return chunk.getDictData()
|
||||
return None
|
||||
|
||||
def setFaceswapDictData (self, dict_data=None):
|
||||
def setDFLDictData (self, dict_data=None):
|
||||
for chunk in self.chunks:
|
||||
if type(chunk) == FaceswapChunk:
|
||||
if type(chunk) == DFLChunk:
|
||||
self.chunks.remove(chunk)
|
||||
break
|
||||
|
||||
if not dict_data is None:
|
||||
chunk = FaceswapChunk(dict_data)
|
||||
chunk = DFLChunk(dict_data)
|
||||
self.chunks.insert(-1, chunk)
|
||||
|
||||
def get_face_type(self):
|
||||
return self.fcwp_dict['face_type']
|
||||
|
||||
def get_landmarks(self):
|
||||
return np.array ( self.fcwp_dict['landmarks'] )
|
||||
|
||||
def get_yaw_value(self):
|
||||
return self.fcwp_dict['yaw_value']
|
||||
|
||||
def get_pitch_value(self):
|
||||
return self.fcwp_dict['pitch_value']
|
||||
|
||||
def get_source_filename(self):
|
||||
return self.fcwp_dict['source_filename']
|
||||
|
||||
def get_source_rect(self):
|
||||
return self.fcwp_dict['source_rect']
|
||||
|
||||
def get_source_landmarks(self):
|
||||
return np.array ( self.fcwp_dict['source_landmarks'] )
|
||||
|
||||
def __str__(self):
|
||||
return "<PNG length={length} chunks={}>".format(len(self.chunks), **self.__dict__)
|
||||
Loading…
Add table
Add a link
Reference in a new issue