mirror of
https://github.com/deepfakes/faceswap
synced 2025-06-09 04:36:50 -04:00
286f6b28b3
- Fix filter/nfilter for convert - Remove existing snapshot folder before creating if it pre-exists - Don't initialize training session until after first save. "Nonetype" session fix
264 lines
11 KiB
Python
264 lines
11 KiB
Python
#!/usr/bin python3
|
|
""" Face and landmarks detection for faceswap.py """
|
|
import logging
|
|
|
|
import numpy as np
|
|
|
|
from lib.aligner import Extract as AlignerExtract, get_align_mat, get_matrix_scaling
|
|
|
|
logger = logging.getLogger(__name__) # pylint: disable=invalid-name
|
|
|
|
|
|
class DetectedFace():
|
|
""" Detected face and landmark information """
|
|
def __init__( # pylint: disable=invalid-name
|
|
self, image=None, x=None, w=None, y=None, h=None,
|
|
landmarksXY=None):
|
|
logger.trace("Initializing %s", self.__class__.__name__)
|
|
self.image = image
|
|
self.x = x
|
|
self.w = w
|
|
self.y = y
|
|
self.h = h
|
|
self.landmarksXY = landmarksXY
|
|
self.hash = None # Hash must be set when the file is saved due to image compression
|
|
|
|
self.aligned = dict()
|
|
self.feed = dict()
|
|
self.reference = dict()
|
|
logger.trace("Initialized %s", self.__class__.__name__)
|
|
|
|
@property
|
|
def extract_ratio(self):
|
|
""" The ratio of padding to add for training images """
|
|
return 0.375
|
|
|
|
@property
|
|
def landmarks_as_xy(self):
|
|
""" Landmarks as XY """
|
|
return self.landmarksXY
|
|
|
|
def to_bounding_box_dict(self):
|
|
""" Return Bounding Box as a bounding box dixt """
|
|
retval = dict(left=self.x, top=self.y, right=self.x + self.w, bottom=self.y + self.h)
|
|
logger.trace("Returning: %s", retval)
|
|
return retval
|
|
|
|
def from_bounding_box_dict(self, bounding_box_dict, image=None):
|
|
""" Set Bounding Box from a bounding box dict """
|
|
logger.trace("Creating from bounding box dict: %s", bounding_box_dict)
|
|
if not isinstance(bounding_box_dict, dict):
|
|
raise ValueError("Supplied Bounding Box is not a dictionary.")
|
|
self.x = bounding_box_dict["left"]
|
|
self.w = bounding_box_dict["right"] - bounding_box_dict["left"]
|
|
self.y = bounding_box_dict["top"]
|
|
self.h = bounding_box_dict["bottom"] - bounding_box_dict["top"]
|
|
if image is not None and image.any():
|
|
self.image_to_face(image)
|
|
logger.trace("Created from bounding box dict: (x: %s, w: %s, y: %s. h: %s)",
|
|
self.x, self.w, self.y, self.h)
|
|
|
|
def image_to_face(self, image):
|
|
""" Crop an image around bounding box to the face
|
|
and capture it's dimensions """
|
|
logger.trace("Cropping face from image")
|
|
self.image = image[self.y: self.y + self.h,
|
|
self.x: self.x + self.w]
|
|
|
|
def to_alignment(self):
|
|
""" Convert a detected face to alignment dict """
|
|
alignment = dict()
|
|
alignment["x"] = self.x
|
|
alignment["w"] = self.w
|
|
alignment["y"] = self.y
|
|
alignment["h"] = self.h
|
|
alignment["landmarksXY"] = self.landmarksXY
|
|
alignment["hash"] = self.hash
|
|
logger.trace("Returning: %s", alignment)
|
|
return alignment
|
|
|
|
def from_alignment(self, alignment, image=None):
|
|
""" Convert a face alignment to detected face object """
|
|
logger.trace("Creating from alignment: (alignment: %s, has_image: %s)",
|
|
alignment, bool(image is not None))
|
|
self.x = alignment["x"]
|
|
self.w = alignment["w"]
|
|
self.y = alignment["y"]
|
|
self.h = alignment["h"]
|
|
self.landmarksXY = alignment["landmarksXY"]
|
|
# Manual tool does not know the final hash so default to None
|
|
self.hash = alignment.get("hash", None)
|
|
if image is not None and image.any():
|
|
self.image_to_face(image)
|
|
logger.trace("Created from alignment: (x: %s, w: %s, y: %s. h: %s, "
|
|
"landmarks: %s)",
|
|
self.x, self.w, self.y, self.h, self.landmarksXY)
|
|
|
|
# <<< Aligned Face methods and properties >>> #
|
|
def load_aligned(self, image, size=256, align_eyes=False, dtype=None):
|
|
""" No need to load aligned information for all uses of this
|
|
class, so only call this to load the information for easy
|
|
reference to aligned properties for this face """
|
|
# Don't reload an already aligned face:
|
|
if self.aligned:
|
|
logger.trace("Skipping alignment calculation for already aligned face")
|
|
else:
|
|
logger.trace("Loading aligned face: (size: %s, align_eyes: %s, dtype: %s)",
|
|
size, align_eyes, dtype)
|
|
padding = int(size * self.extract_ratio) // 2
|
|
self.aligned["size"] = size
|
|
self.aligned["padding"] = padding
|
|
self.aligned["align_eyes"] = align_eyes
|
|
self.aligned["matrix"] = get_align_mat(self, size, align_eyes)
|
|
self.aligned["face"] = None
|
|
if image is not None and self.aligned["face"] is None:
|
|
logger.trace("Getting aligned face")
|
|
face = AlignerExtract().transform(
|
|
image,
|
|
self.aligned["matrix"],
|
|
size,
|
|
padding)
|
|
self.aligned["face"] = face if dtype is None else face.astype(dtype)
|
|
|
|
logger.trace("Loaded aligned face: %s", {key: val
|
|
for key, val in self.aligned.items()
|
|
if key != "face"})
|
|
|
|
def padding_from_coverage(self, size, coverage_ratio):
|
|
""" Return the image padding for a face from coverage_ratio set against a
|
|
pre-padded training image """
|
|
adjusted_ratio = coverage_ratio - (1 - self.extract_ratio)
|
|
padding = round((size * adjusted_ratio) / 2)
|
|
logger.trace(padding)
|
|
return padding
|
|
|
|
def load_feed_face(self, image, size=64, coverage_ratio=0.625, dtype=None):
|
|
""" Return a face in the correct dimensions for feeding into a NN
|
|
|
|
Coverage ratio should be the ratio of the extracted image that was used for
|
|
training """
|
|
logger.trace("Loading feed face: (size: %s, coverage_ratio: %s, dtype: %s)",
|
|
size, coverage_ratio, dtype)
|
|
|
|
self.feed["size"] = size
|
|
self.feed["padding"] = self.padding_from_coverage(size, coverage_ratio)
|
|
self.feed["matrix"] = get_align_mat(self, size, should_align_eyes=False)
|
|
|
|
face = np.clip(AlignerExtract().transform(image,
|
|
self.feed["matrix"],
|
|
size,
|
|
self.feed["padding"])[:, :, :3] / 255.0,
|
|
0.0, 1.0)
|
|
self.feed["face"] = face if dtype is None else face.astype(dtype)
|
|
|
|
logger.trace("Loaded feed face. (face_shape: %s, matrix: %s)",
|
|
self.feed_face.shape, self.feed_matrix)
|
|
|
|
def load_reference_face(self, image, size=64, coverage_ratio=0.625, dtype=None):
|
|
""" Return a face in the correct dimensions for reference to the output from a NN
|
|
|
|
Coverage ratio should be the ratio of the extracted image that was used for
|
|
training """
|
|
logger.trace("Loading reference face: (size: %s, coverage_ratio: %s, dtype: %s)",
|
|
size, coverage_ratio, dtype)
|
|
|
|
self.reference["size"] = size
|
|
self.reference["padding"] = self.padding_from_coverage(size, coverage_ratio)
|
|
self.reference["matrix"] = get_align_mat(self, size, should_align_eyes=False)
|
|
|
|
face = np.clip(AlignerExtract().transform(image,
|
|
self.reference["matrix"],
|
|
size,
|
|
self.reference["padding"])[:, :, :3] / 255.0,
|
|
0.0, 1.0)
|
|
self.reference["face"] = face if dtype is None else face.astype(dtype)
|
|
|
|
logger.trace("Loaded reference face. (face_shape: %s, matrix: %s)",
|
|
self.reference_face.shape, self.reference_matrix)
|
|
|
|
@property
|
|
def original_roi(self):
|
|
""" Return the square aligned box location on the original
|
|
image """
|
|
roi = AlignerExtract().get_original_roi(self.aligned["matrix"],
|
|
self.aligned["size"],
|
|
self.aligned["padding"])
|
|
logger.trace("Returning: %s", roi)
|
|
return roi
|
|
|
|
@property
|
|
def aligned_landmarks(self):
|
|
""" Return the landmarks location transposed to extracted face """
|
|
landmarks = AlignerExtract().transform_points(self.landmarksXY,
|
|
self.aligned["matrix"],
|
|
self.aligned["size"],
|
|
self.aligned["padding"])
|
|
logger.trace("Returning: %s", landmarks)
|
|
return landmarks
|
|
|
|
@property
|
|
def aligned_face(self):
|
|
""" Return aligned detected face """
|
|
return self.aligned["face"]
|
|
|
|
@property
|
|
def adjusted_matrix(self):
|
|
""" Return adjusted matrix for size/padding combination """
|
|
mat = AlignerExtract().transform_matrix(self.aligned["matrix"],
|
|
self.aligned["size"],
|
|
self.aligned["padding"])
|
|
logger.trace("Returning: %s", mat)
|
|
return mat
|
|
|
|
@property
|
|
def adjusted_interpolators(self):
|
|
""" Return the interpolator and reverse interpolator for the adjusted matrix """
|
|
return get_matrix_scaling(self.adjusted_matrix)
|
|
|
|
@property
|
|
def feed_face(self):
|
|
""" Return face for feeding into NN """
|
|
return self.feed["face"]
|
|
|
|
@property
|
|
def feed_matrix(self):
|
|
""" Return matrix for transforming feed face back to image """
|
|
mat = AlignerExtract().transform_matrix(self.feed["matrix"],
|
|
self.feed["size"],
|
|
self.feed["padding"])
|
|
logger.trace("Returning: %s", mat)
|
|
return mat
|
|
|
|
@property
|
|
def feed_interpolators(self):
|
|
""" Return the interpolators for an input face """
|
|
return get_matrix_scaling(self.feed_matrix)
|
|
|
|
@property
|
|
def reference_face(self):
|
|
""" Return source face at size of output from NN for reference """
|
|
return self.reference["face"]
|
|
|
|
@property
|
|
def reference_landmarks(self):
|
|
""" Return the landmarks location transposed to reference face """
|
|
landmarks = AlignerExtract().transform_points(self.landmarksXY,
|
|
self.reference["matrix"],
|
|
self.reference["size"],
|
|
self.reference["padding"])
|
|
logger.trace("Returning: %s", landmarks)
|
|
return landmarks
|
|
|
|
@property
|
|
def reference_matrix(self):
|
|
""" Return matrix for transforming output face back to image """
|
|
mat = AlignerExtract().transform_matrix(self.reference["matrix"],
|
|
self.reference["size"],
|
|
self.reference["padding"])
|
|
logger.trace("Returning: %s", mat)
|
|
return mat
|
|
|
|
@property
|
|
def reference_interpolators(self):
|
|
""" Return the interpolators for an output face """
|
|
return get_matrix_scaling(self.reference_matrix)
|