Move code to face helper

facefusion/face_helper.py

@@ -0,0 +1,52 @@
+from typing import Tuple, Dict
+
+import cv2
+import numpy
+
+from facefusion.typing import Face, Frame, Matrix, Template
+
+TEMPLATES : Dict[str, numpy.array] =\
+{
+	'arface': numpy.array(
+	[
+		[ 38.2946, 51.6963 ],
+		[ 73.5318, 51.5014 ],
+		[ 56.0252, 71.7366 ],
+		[ 41.5493, 92.3655 ],
+		[ 70.7299, 92.2041 ]
+	]),
+	'ffhq': numpy.array(
+	[
+		[ 192.98138, 239.94708 ],
+		[ 318.90277, 240.1936 ],
+		[ 256.63416, 314.01935 ],
+		[ 201.26117, 371.41043 ],
+		[ 313.08905, 371.15118 ]
+	])
+}
+
+
+def warp_face(target_face : Face, temp_frame : Frame, template : Template) -> Tuple[Frame, Matrix]:
+	affine_matrix = cv2.estimateAffinePartial2D(target_face.kps, TEMPLATES[template], method = cv2.LMEDS)[0]
+	crop_frame = cv2.warpAffine(temp_frame, affine_matrix, (512, 512))
+	return crop_frame, affine_matrix
+
+
+def paste_back(temp_frame : Frame, crop_frame : Frame, affine_matrix : Matrix) -> Frame:
+	inverse_affine_matrix = cv2.invertAffineTransform(affine_matrix)
+	temp_frame_height, temp_frame_width = temp_frame.shape[0:2]
+	crop_frame_height, crop_frame_width = crop_frame.shape[0:2]
+	inverse_crop_frame = cv2.warpAffine(crop_frame, inverse_affine_matrix, (temp_frame_width, temp_frame_height))
+	inverse_mask = numpy.ones((crop_frame_height, crop_frame_width, 3), dtype = numpy.float32)
+	inverse_mask_frame = cv2.warpAffine(inverse_mask, inverse_affine_matrix, (temp_frame_width, temp_frame_height))
+	inverse_mask_frame = cv2.erode(inverse_mask_frame, numpy.ones((2, 2)))
+	inverse_mask_border = inverse_mask_frame * inverse_crop_frame
+	inverse_mask_area = numpy.sum(inverse_mask_frame) // 3
+	inverse_mask_edge = int(inverse_mask_area ** 0.5) // 20
+	inverse_mask_radius = inverse_mask_edge * 2
+	inverse_mask_center = cv2.erode(inverse_mask_frame, numpy.ones((inverse_mask_radius, inverse_mask_radius)))
+	inverse_mask_blur_size = inverse_mask_edge * 2 + 1
+	inverse_mask_blur_area = cv2.GaussianBlur(inverse_mask_center, (inverse_mask_blur_size, inverse_mask_blur_size), 0)
+	temp_frame = inverse_mask_blur_area * inverse_mask_border + (1 - inverse_mask_blur_area) * temp_frame
+	temp_frame = temp_frame.clip(0, 255).astype(numpy.uint8)
+	return temp_frame

facefusion/installer.py

@@ -46,16 +46,8 @@ def run(program : ArgumentParser) -> None:
 	else:
 		answers = inquirer.prompt(
 		[
-			inquirer.List(
+			inquirer.List('torch', message = wording.get('install_dependency_help').format(dependency = 'torch'), choices = list(TORCH.keys())),
-				'torch',
+			inquirer.List('onnxruntime', message = wording.get('install_dependency_help').format(dependency = 'onnxruntime'), choices = list(ONNXRUNTIMES.keys()))
-				message = wording.get('install_dependency_help').format(dependency = 'torch'),
-				choices = list(TORCH.keys())
-			),
-			inquirer.List(
-				'onnxruntime',
-				message = wording.get('install_dependency_help').format(dependency = 'onnxruntime'),
-				choices = list(ONNXRUNTIMES.keys())
-			)
 		])
 	if answers is not None:
 		torch = answers['torch']

facefusion/processors/frame/modules/face_enhancer.py

@@ -1,4 +1,4 @@
-from typing import Any, List, Tuple, Dict, Literal, Optional
+from typing import Any, List, Dict, Literal, Optional
 from argparse import ArgumentParser
 import cv2
 import threading
@@ -9,7 +9,8 @@ import facefusion.globals
 from facefusion import wording
 from facefusion.core import update_status
 from facefusion.face_analyser import get_many_faces, clear_face_analyser
-from facefusion.typing import Face, Frame, Matrix, Update_Process, ProcessMode, ModelValue, OptionsWithModel
+from facefusion.face_helper import warp_face, paste_back
+from facefusion.typing import Face, Frame, Update_Process, ProcessMode, ModelValue, OptionsWithModel
 from facefusion.utilities import conditional_download, resolve_relative_path, is_image, is_video, is_file, is_download_done
 from facefusion.vision import read_image, read_static_image, write_image
 from facefusion.processors.frame import globals as frame_processors_globals
@@ -128,7 +129,7 @@ def post_process() -> None:
 
 def enhance_face(target_face: Face, temp_frame: Frame) -> Frame:
 	frame_processor = get_frame_processor()
-	crop_frame, affine_matrix = warp_face(target_face, temp_frame)
+	crop_frame, affine_matrix = warp_face(target_face, temp_frame, 'ffhq')
 	crop_frame = prepare_crop_frame(crop_frame)
 	frame_processor_inputs = {}
 	for frame_processor_input in frame_processor.get_inputs():
@@ -144,20 +145,6 @@ def enhance_face(target_face: Face, temp_frame: Frame) -> Frame:
 	return temp_frame
 
 
-def warp_face(target_face : Face, temp_frame : Frame) -> Tuple[Frame, Matrix]:
-	template = numpy.array(
-	[
-		[ 192.98138, 239.94708 ],
-		[ 318.90277, 240.1936 ],
-		[ 256.63416, 314.01935 ],
-		[ 201.26117, 371.41043 ],
-		[ 313.08905, 371.15118 ]
-	])
-	affine_matrix = cv2.estimateAffinePartial2D(target_face['kps'], template, method = cv2.LMEDS)[0]
-	crop_frame = cv2.warpAffine(temp_frame, affine_matrix, (512, 512))
-	return crop_frame, affine_matrix
-
-
 def prepare_crop_frame(crop_frame : Frame) -> Frame:
 	crop_frame = crop_frame[:, :, ::-1] / 255.0
 	crop_frame = (crop_frame - 0.5) / 0.5
@@ -174,26 +161,6 @@ def normalize_crop_frame(crop_frame : Frame) -> Frame:
 	return crop_frame
 
 
-def paste_back(temp_frame : Frame, crop_frame : Frame, affine_matrix : Matrix) -> Frame:
-	inverse_affine_matrix = cv2.invertAffineTransform(affine_matrix)
-	temp_frame_height, temp_frame_width = temp_frame.shape[0:2]
-	crop_frame_height, crop_frame_width = crop_frame.shape[0:2]
-	inverse_crop_frame = cv2.warpAffine(crop_frame, inverse_affine_matrix, (temp_frame_width, temp_frame_height))
-	inverse_mask = numpy.ones((crop_frame_height, crop_frame_width, 3), dtype = numpy.float32)
-	inverse_mask_frame = cv2.warpAffine(inverse_mask, inverse_affine_matrix, (temp_frame_width, temp_frame_height))
-	inverse_mask_frame = cv2.erode(inverse_mask_frame, numpy.ones((2, 2)))
-	inverse_mask_border = inverse_mask_frame * inverse_crop_frame
-	inverse_mask_area = numpy.sum(inverse_mask_frame) // 3
-	inverse_mask_edge = int(inverse_mask_area ** 0.5) // 20
-	inverse_mask_radius = inverse_mask_edge * 2
-	inverse_mask_center = cv2.erode(inverse_mask_frame, numpy.ones((inverse_mask_radius, inverse_mask_radius)))
-	inverse_mask_blur_size = inverse_mask_edge * 2 + 1
-	inverse_mask_blur_area = cv2.GaussianBlur(inverse_mask_center, (inverse_mask_blur_size, inverse_mask_blur_size), 0)
-	temp_frame = inverse_mask_blur_area * inverse_mask_border + (1 - inverse_mask_blur_area) * temp_frame
-	temp_frame = temp_frame.clip(0, 255).astype(numpy.uint8)
-	return temp_frame
-
-
 def blend_frame(temp_frame : Frame, paste_frame : Frame) -> Frame:
 	face_enhancer_blend = 1 - (frame_processors_globals.face_enhancer_blend / 100)
 	temp_frame = cv2.addWeighted(temp_frame, face_enhancer_blend, paste_frame, 1 - face_enhancer_blend, 0)

facefusion/typing.py

@@ -9,6 +9,7 @@ Matrix = numpy.ndarray[Any, Any]
 Update_Process = Callable[[], None]
 Process_Frames = Callable[[str, List[str], Update_Process], None]
 
+Template = Literal[ 'arcface', 'ffhq' ]
 ProcessMode = Literal[ 'output', 'preview', 'stream' ]
 FaceRecognition = Literal[ 'reference', 'many' ]
 FaceAnalyserDirection = Literal[ 'left-right', 'right-left', 'top-bottom', 'bottom-top', 'small-large', 'large-small' ]