Recrosoftware/sd-webui-reactor
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Added Masking Tab

Browse Source 
Added Masking tab to allow further configuration of the masking feature. Added ability to save the masked face as a separate file with alpha transparency for easy editing in photoshop in a layer over the original
This commit is contained in:
jiveabillion 2023-11-27 12:23:44 -05:00
parent 87cd569324
commit de84e022ee
4 changed files with 213 additions and 53 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
scripts/inferencers/bisenet_mask_generator.py
View File

@ -4,16 +4,19 @@ import cv2
import modules.shared as shared
import numpy as np
import torch
from scripts.reactor_logger import logger
from facexlib.parsing import init_parsing_model
from facexlib.utils.misc import img2tensor
from torchvision.transforms.functional import normalize
from PIL import Image
from scripts.inferencers.vignette_mask_generator import VignetteMaskGenerator
from scripts.inferencers.mask_generator import MaskGenerator
from scripts.reactor_logger import logger
class BiSeNetMaskGenerator(MaskGenerator):
def __init__(self) -> None:
self.mask_model = init_parsing_model(device=shared.device)
self.fallback_mask_generator = VignetteMaskGenerator()
def name(self):
return "BiSeNet"
@ -25,7 +28,7 @@ class BiSeNetMaskGenerator(MaskGenerator):
affected_areas: List[str],
mask_size: int,
use_minimal_area: bool,
fallback_ratio: float = 0.25,
fallback_ratio: float = 0.10,
**kwargs,
) -> np.ndarray:
original_face_image = face_image
@ -59,11 +62,11 @@ class BiSeNetMaskGenerator(MaskGenerator):
if w != 512 or h != 512:
mask = cv2.resize(mask, dsize=(w, h))
"""if MaskGenerator.calculate_mask_coverage(mask) < fallback_ratio:
logger.info("Use fallback mask generator")
if MaskGenerator.calculate_mask_coverage(mask) < fallback_ratio:
logger.status(F"Mask coverage less than fallback ratio of {fallback_ratio}. Using vignette mask generator.")
mask = self.fallback_mask_generator.generate_mask(
original_face_image, face_area_on_image, use_minimal_area=True
)"""
)
return mask

50
scripts/inferencers/vignette_mask_generator.py Normal file
View File

@ -0,0 +1,50 @@
from typing import Tuple
import cv2
import numpy as np
from scripts.inferencers.mask_generator import MaskGenerator
class VignetteMaskGenerator(MaskGenerator):
def name(self):
return "Vignette"
def generate_mask(
self,
face_image: np.ndarray,
face_area_on_image: Tuple[int, int, int, int],
use_minimal_area: bool,
sigma: float = -1,
keep_safe_area: bool = False,
**kwargs,
) -> np.ndarray:
(left, top, right, bottom) = face_area_on_image
w, h = right - left, bottom - top
mask = np.zeros((face_image.shape[0], face_image.shape[1]), dtype=np.uint8)
if use_minimal_area:
sigma = 120 if sigma == -1 else sigma
mask[top : top + h, left : left + w] = 255
else:
sigma = 180 if sigma == -1 else sigma
h, w = face_image.shape[0], face_image.shape[1]
mask[:, :] = 255
Y = np.linspace(0, h, h, endpoint=False)
X = np.linspace(0, w, w, endpoint=False)
Y, X = np.meshgrid(Y, X)
Y -= h / 2
X -= w / 2
gaussian = np.exp(-(X**2 + Y**2) / (2 * sigma**2))
gaussian_mask = np.uint8(255 * gaussian.T)
if use_minimal_area:
mask[top : top + h, left : left + w] = gaussian_mask
else:
mask[:, :] = gaussian_mask
if keep_safe_area:
mask = cv2.ellipse(mask, ((left + right) // 2, (top + bottom) // 2), (w // 2, h // 2), 0, 0, 360, 255, -1)
mask = cv2.cvtColor(mask, cv2.COLOR_GRAY2RGB)
return mask

133
scripts/reactor_faceswap.py
View File

@ -28,13 +28,14 @@ except:
model_path = os.path.abspath("models")
from scripts.reactor_logger import logger
from scripts.reactor_swapper import EnhancementOptions, swap_face, check_process_halt, reset_messaged
from scripts.reactor_swapper import EnhancementOptions,MaskOptions,MaskOption, swap_face, check_process_halt, reset_messaged
from scripts.reactor_version import version_flag, app_title
from scripts.console_log_patch import apply_logging_patch
from scripts.reactor_helpers import make_grid, get_image_path, set_Device
from scripts.reactor_globals import DEVICE, DEVICE_LIST
MODELS_PATH = None
def get_models():
@ -66,8 +67,8 @@ class FaceSwapScript(scripts.Script):
img = gr.Image(type="pil")
enable = gr.Checkbox(False, label="Enable", info=f"The Fast and Simple FaceSwap Extension - {version_flag}")
save_original = gr.Checkbox(False, label="Save Original", info="Save the original image(s) made before swapping; If you use \"img2img\" - this option will affect with \"Swap in generated\" only")
mask_face = gr.Checkbox(False, label="Mask Faces", info="Attempt to mask only the faces and eliminate pixelation of the image around the contours.")
mask_face = gr.Checkbox(False, label="Mask Faces", info="Attempt to mask only the faces and eliminate pixelation of the image around the contours. Additional settings in the Masking tab.")
gr.Markdown("<br>")
gr.Markdown("Source Image (above):")
with gr.Row():
@ -140,6 +141,24 @@ class FaceSwapScript(scripts.Script):
upscaler_visibility = gr.Slider(
0, 1, 1, step=0.1, label="Upscaler Visibility (if scale = 1)"
)
with gr.Tab("Masking"):
save_face_mask = gr.Checkbox(False, label="Save Face Mask", info="Save the face mask as a separate image with alpha transparency.")
mask_areas = gr.CheckboxGroup(
label="Mask areas", choices=["Face", "Hair", "Hat", "Neck"], type="value", value= MaskOption.DEFAULT_FACE_AREAS
)
mask_blur = gr.Slider(label="Mask blur ", minimum=0, maximum=64, step=1, value=MaskOption.DEFAULT_MASK_BLUR,info="The number of pixels from the outer edge of the mask to blur.")
face_size = gr.Radio(
label = "Face Size", choices = [512,256,128],value=MaskOption.DEFAULT_FACE_SIZE,type="value", info="Size of the masked area. Use larger numbers if the face is expected to be large, smaller if small. Default is 512."
)
mask_vignette_fallback_threshold = gr.Slider(
minimum=0.1,
maximum=1.0,
step=0.01,
value=MaskOption.DEFAULT_VIGNETTE_THRESHOLD,
label="Vignette fallback threshold",
info="Switch to a rectangular vignette mask when masked area is only this specified percentage of Face Size."
)
use_minimal_area = gr.Checkbox(MaskOption.DEFAULT_USE_MINIMAL_AREA, label="Use Minimal Area", info="Use the least amount of area for the mask as possible. This is good for multiple faces that are close together or for preserving the most of the surrounding image.")
with gr.Tab("Settings"):
models = get_models()
with gr.Row(visible=EP_is_visible):
@ -213,7 +232,14 @@ class FaceSwapScript(scripts.Script):
source_hash_check,
target_hash_check,
device,
mask_face
mask_face,
save_face_mask,
mask_areas,
mask_blur,
use_minimal_area,
face_size,
mask_vignette_fallback_threshold,
]
@ -242,7 +268,17 @@ class FaceSwapScript(scripts.Script):
restorer_visibility=self.face_restorer_visibility,
codeformer_weight=self.codeformer_weight,
)
@property
def mask_options(self) -> MaskOptions:
return MaskOptions(
mask_areas = self.mask_areas,
save_face_mask = self.save_face_mask,
mask_blur = self.mask_blur,
face_size = self.mask_face_size,
vignette_fallback_threshold = self.mask_vignette_fallback_threshold,
use_minimal_area = self.mask_use_minimal_area,
)
def process(
self,
p: StableDiffusionProcessing,
@ -267,7 +303,14 @@ class FaceSwapScript(scripts.Script):
source_hash_check,
target_hash_check,
device,
mask_face
mask_face,
save_face_mask,
mask_areas,
mask_blur,
mask_use_minimal_area,
mask_face_size,
mask_vignette_fallback_threshold,
):
self.enable = enable
if self.enable:
@ -296,6 +339,12 @@ class FaceSwapScript(scripts.Script):
self.target_hash_check = target_hash_check
self.device = device
self.mask_face = mask_face
self.save_face_mask = save_face_mask
self.mask_blur = mask_blur
self.mask_areas = mask_areas
self.mask_face_size = mask_face_size
self.mask_vignette_fallback_threshold = mask_vignette_fallback_threshold
self.mask_use_minimal_area = mask_use_minimal_area
if self.gender_source is None or self.gender_source == "No":
self.gender_source = 0
if self.gender_target is None or self.gender_target == "No":
@ -318,9 +367,10 @@ class FaceSwapScript(scripts.Script):
self.target_hash_check = False
set_Device(self.device)
logger.status(f"Self: {self}")
if self.source is not None:
apply_logging_patch(console_logging_level)
if isinstance(p, StableDiffusionProcessingImg2Img) and self.swap_in_source:
logger.status("Working: source face index %s, target face index %s", self.source_faces_index, self.faces_index)
@ -339,7 +389,8 @@ class FaceSwapScript(scripts.Script):
source_hash_check=self.source_hash_check,
target_hash_check=self.target_hash_check,
device=self.device,
mask_face=mask_face
mask_face=mask_face,
mask_options=self.mask_options
)
p.init_images[i] = result
# result_path = get_image_path(p.init_images[i], p.outpath_samples, "", p.all_seeds[i], p.all_prompts[i], "txt", p=p, suffix="-swapped")
@ -373,13 +424,14 @@ class FaceSwapScript(scripts.Script):
if self.swap_in_generated:
logger.status("Working: source face index %s, target face index %s", self.source_faces_index, self.faces_index)
if self.source is not None:
for i,(img,info) in enumerate(zip(orig_images, orig_infotexts)):
if check_process_halt():
postprocess_run = False
break
if len(orig_images) > 1:
logger.status("Swap in %s", i)
result, output, swapped = swap_face(
result, output, swapped, masked_faces = swap_face(
self.source,
img,
source_faces_index=self.source_faces_index,
@ -391,7 +443,8 @@ class FaceSwapScript(scripts.Script):
source_hash_check=self.source_hash_check,
target_hash_check=self.target_hash_check,
device=self.device,
mask_face=self.mask_face
mask_face=self.mask_face,
mask_options=self.mask_options
)
if result is not None and swapped > 0:
result_images.append(result)
@ -400,6 +453,14 @@ class FaceSwapScript(scripts.Script):
img_path = save_image(result, p.outpath_samples, "", p.all_seeds[0], p.all_prompts[0], "png",info=info, p=p, suffix=suffix)
except:
logger.error("Cannot save a result image - please, check SD WebUI Settings (Saving and Paths)")
if self.mask_face and self.save_face_mask and masked_faces is not None:
result_images.append(masked_faces)
suffix = "-mask"
try:
img_path = save_image(masked_faces, p.outpath_samples, "", p.all_seeds[0], p.all_prompts[0], "png",info=info, p=p, suffix=suffix)
except:
logger.error("Cannot save a Masked Face image - please, check SD WebUI Settings (Saving and Paths)")
elif result is None:
logger.error("Cannot create a result image")
@ -476,7 +537,7 @@ class FaceSwapScriptExtras(scripts_postprocessing.ScriptPostprocessing):
with gr.Column():
img = gr.Image(type="pil")
enable = gr.Checkbox(False, label="Enable", info=f"The Fast and Simple FaceSwap Extension - {version_flag}")
mask_face = gr.Checkbox(False, label="Mask Faces", info="Attempt to mask only the faces and eliminate pixelation of the image around the contours.")
mask_face = gr.Checkbox(False, label="Mask Faces", info="Attempt to mask only the faces and eliminate pixelation of the image around the contours. Additional settings in the Masking tab.")
gr.Markdown("Source Image (above):")
with gr.Row():
@ -536,6 +597,25 @@ class FaceSwapScriptExtras(scripts_postprocessing.ScriptPostprocessing):
upscaler_visibility = gr.Slider(
0, 1, 1, step=0.1, label="Upscaler Visibility (if scale = 1)"
)
with gr.Tab("Masking"):
save_face_mask = gr.Checkbox(False, label="Save Face Mask", info="Save the face mask as a separate image with alpha transparency.")
mask_areas = gr.CheckboxGroup(
label="Mask areas", choices=["Face", "Hair", "Hat", "Neck"], type="value", value= MaskOption.DEFAULT_FACE_AREAS
)
mask_blur = gr.Slider(label="Mask blur ", minimum=0, maximum=64, step=1, value=MaskOption.DEFAULT_MASK_BLUR,info="The number of pixels from the outer edge of the mask to blur.")
face_size = gr.Radio(
label = "Face Size", choices = [512,256,128],value=MaskOption.DEFAULT_FACE_SIZE,type="value", info="Size of the masked area. Use larger numbers if the face is expected to be large, smaller if small. Default is 512."
)
mask_vignette_fallback_threshold = gr.Slider(
minimum=0.1,
maximum=1.0,
step=0.01,
value=MaskOption.DEFAULT_VIGNETTE_THRESHOLD,
label="Vignette fallback threshold",
info="Switch to a rectangular vignette mask when masked area is only this specified percentage of Face Size."
)
use_minimal_area = gr.Checkbox(MaskOption.DEFAULT_USE_MINIMAL_AREA, label="Use Minimal Area", info="Use the least amount of area for the mask as possible. This is good for multiple faces that are close together or for preserving the most of the surrounding image.")
with gr.Tab("Settings"):
models = get_models()
with gr.Row(visible=EP_is_visible):
@ -592,7 +672,13 @@ class FaceSwapScriptExtras(scripts_postprocessing.ScriptPostprocessing):
'gender_target': gender_target,
'codeformer_weight': codeformer_weight,
'device': device,
'mask_face':mask_face
'mask_face':mask_face,
'save_face_mask':save_face_mask,
'mask_areas':mask_areas,
'mask_blur':mask_blur,
'mask_vignette_fallback_threshold':mask_vignette_fallback_threshold,
'face_size':face_size,
'use_minimal_area':use_minimal_area,
}
return args
@ -621,13 +707,21 @@ class FaceSwapScriptExtras(scripts_postprocessing.ScriptPostprocessing):
restorer_visibility=self.face_restorer_visibility,
codeformer_weight=self.codeformer_weight,
)
@property
def mask_options(self) -> MaskOptions:
return MaskOptions(
mask_areas = self.mask_areas,
save_face_mask = self.save_face_mask,
mask_blur = self.mask_blur,
face_size = self.mask_face_size,
vignette_fallback_threshold = self.mask_vignette_fallback_threshold,
use_minimal_area = self.mask_use_minimal_area,
)
def process(self, pp: scripts_postprocessing.PostprocessedImage, **args):
if args['enable']:
reset_messaged()
if check_process_halt():
return
return
global MODELS_PATH
self.source = args['img']
self.face_restorer_name = args['face_restorer_name']
@ -643,6 +737,12 @@ class FaceSwapScriptExtras(scripts_postprocessing.ScriptPostprocessing):
self.codeformer_weight = args['codeformer_weight']
self.device = args['device']
self.mask_face = args['mask_face']
self.save_face_mask = args['save_face_mask']
self.mask_areas= args['mask_areas']
self.mask_blur= args['mask_blur']
self.mask_vignette_fallback_threshold= args['mask_vignette_fallback_threshold']
self.face_size= args['face_size']
self.use_minimal_area= args['use_minimal_area']
if self.gender_source is None or self.gender_source == "No":
self.gender_source = 0
if self.gender_target is None or self.gender_target == "No":
@ -681,7 +781,8 @@ class FaceSwapScriptExtras(scripts_postprocessing.ScriptPostprocessing):
source_hash_check=True,
target_hash_check=True,
device=self.device,
mask_face=self.mask_face
mask_face=self.mask_face,
mask_options=self.mask_options
)
try:
pp.info["ReActor"] = True

70
scripts/reactor_swapper.py
View File

@ -1,7 +1,7 @@
import copy
import os
from dataclasses import dataclass
from typing import List, Union
from typing import List, Tuple, Union
import cv2
import numpy as np
@ -41,7 +41,12 @@ if DEVICE == "CUDA":
PROVIDERS = ["CUDAExecutionProvider"]
else:
PROVIDERS = ["CPUExecutionProvider"]
class MaskOption:
DEFAULT_FACE_AREAS = ["Face"]
DEFAULT_FACE_SIZE = 512
DEFAULT_VIGNETTE_THRESHOLD = 0.1
DEFAULT_MASK_BLUR = 12,
DEFAULT_USE_MINIMAL_AREA = True
@dataclass
class EnhancementOptions:
@ -52,7 +57,16 @@ class EnhancementOptions:
face_restorer: FaceRestoration = None
restorer_visibility: float = 0.5
codeformer_weight: float = 0.5
@dataclass
class MaskOptions:
mask_areas:List[str]
save_face_mask: bool = False
mask_blur:int = 12
face_size:int = 512
vignette_fallback_threshold:float =0.10
use_minimal_area:bool = True
MESSAGED_STOPPED = False
MESSAGED_SKIPPED = False
@ -175,7 +189,7 @@ def enhance_image(image: Image, enhancement_options: EnhancementOptions):
result_image = restore_face(result_image, enhancement_options)
return result_image
def enhance_image_and_mask(image: Image.Image, enhancement_options: EnhancementOptions,target_img_orig:Image.Image,entire_mask_image:Image.Image)->Image.Image:
def enhance_image_and_mask(image: Image.Image, enhancement_options: EnhancementOptions,target_img_orig:Image.Image,entire_mask_image:Image.Image)->Tuple[Image.Image,Image.Image]:
result_image = image
if check_process_halt(msgforced=True):
@ -183,7 +197,11 @@ def enhance_image_and_mask(image: Image.Image, enhancement_options: EnhancementO
if enhancement_options.do_restore_first:
result_image = restore_face(result_image, enhancement_options)
transparent = Image.new("RGBA",result_image.size)
masked_faces = Image.composite(result_image.convert("RGBA"),transparent,entire_mask_image)
result_image = Image.composite(result_image,target_img_orig,entire_mask_image)
result_image = upscale_image(result_image, enhancement_options)
@ -191,10 +209,13 @@ def enhance_image_and_mask(image: Image.Image, enhancement_options: EnhancementO
result_image = upscale_image(result_image, enhancement_options)
entire_mask_image = Image.fromarray(cv2.resize(np.array(entire_mask_image),result_image.size, interpolation=cv2.INTER_AREA)).convert("L")
result_image = Image.composite(result_image,target_img_orig,entire_mask_image)
result_image = restore_face(result_image, enhancement_options)
transparent = Image.new("RGBA",result_image.size)
masked_faces = Image.composite(result_image.convert("RGBA"),transparent,entire_mask_image)
return result_image, masked_faces
return result_image
def get_gender(face, face_index):
@ -310,7 +331,8 @@ def swap_face(
source_hash_check: bool = True,
target_hash_check: bool = False,
device: str = "CPU",
mask_face:bool = False
mask_face:bool = False,
mask_options:Union[MaskOptions, None]= None
):
global SOURCE_FACES, SOURCE_IMAGE_HASH, TARGET_FACES, TARGET_IMAGE_HASH, PROVIDERS
result_image = target_img
@ -444,7 +466,7 @@ def swap_face(
swapped_image = face_swapper.get(result, target_face, source_face)
if mask_face:
result = apply_face_mask(swapped_image=swapped_image,target_image=result,target_face=target_face,entire_mask_image=entire_mask_image)
result = apply_face_mask(swapped_image=swapped_image,target_image=result,target_face=target_face,entire_mask_image=entire_mask_image,mask_options=mask_options)
else:
result = swapped_image
swapped += 1
@ -480,8 +502,8 @@ def swap_face(
result_image = Image.fromarray(cv2.cvtColor(result, cv2.COLOR_BGR2RGB))
if enhancement_options is not None and swapped > 0:
if mask_face and entire_mask_image is not None:
result_image = enhance_image_and_mask(result_image, enhancement_options,Image.fromarray(target_img_orig),Image.fromarray(entire_mask_image).convert("L"))
if mask_face and entire_mask_image is not None:
result_image, masked_faces = enhance_image_and_mask(result_image, enhancement_options,Image.fromarray(target_img_orig),Image.fromarray(entire_mask_image).convert("L"))
else:
result_image = enhance_image(result_image, enhancement_options)
elif mask_face and entire_mask_image is not None and swapped > 0:
@ -492,20 +514,19 @@ def swap_face(
else:
logger.status("No source face(s) found")
return result_image, output, swapped
return result_image, output, swapped,masked_faces
def apply_face_mask(swapped_image:np.ndarray,target_image:np.ndarray,target_face,entire_mask_image:np.array)->np.ndarray:
logger.status("Masking Face")
def apply_face_mask(swapped_image:np.ndarray,target_image:np.ndarray,target_face,entire_mask_image:np.array,mask_options:Union[MaskOptions,None] = None)->np.ndarray:
logger.status("Masking Face")
mask_generator = BiSeNetMaskGenerator()
face = Face(target_image,Rect.from_ndarray(np.array(target_face.bbox)),1.6,512,"")
face = Face(target_image,Rect.from_ndarray(np.array(target_face.bbox)),1.6,mask_options.face_size,"")
face_image = np.array(face.image)
process_face_image(face)
face_area_on_image = face.face_area_on_image
mask = mask_generator.generate_mask(face_image,face_area_on_image=face_area_on_image,affected_areas=["Face"],mask_size=0,use_minimal_area=True)
mask = cv2.blur(mask, (12, 12))
"""entire_mask_image = np.zeros_like(target_image)"""
mask = mask_generator.generate_mask(face_image,face_area_on_image=face_area_on_image,affected_areas=mask_options.mask_areas,mask_size=0,use_minimal_area=mask_options.use_minimal_area)
mask = cv2.blur(mask, (mask_options.mask_blur, mask_options.mask_blur))
larger_mask = cv2.resize(mask, dsize=(face.width, face.height))
entire_mask_image[
face.top : face.bottom,
@ -554,22 +575,7 @@ def color_generator(colors):
color_iter = color_generator(colors)
def process_face_image(
face: Face,
**kwargs,
) -> Image:
image = np.array(face.image)
overlay = image.copy()
cv2.rectangle(overlay, (0, 0), (image.shape[1], image.shape[0]), next(color_iter), -1)
l, t, r, b = face.face_area_on_image
cv2.rectangle(overlay, (l, t), (r, b), (0, 0, 0), 10)
if face.landmarks_on_image is not None:
for landmark in face.landmarks_on_image:
cv2.circle(overlay, (int(landmark.x), int(landmark.y)), 6, (0, 0, 0), 10)
alpha = 0.3
output = cv2.addWeighted(image, 1 - alpha, overlay, alpha, 0)
return Image.fromarray(output)
def dilate_erode(img: Image.Image, value: int) -> Image.Image:
"""
The dilate_erode function takes an image and a value.