facefusion/facefusion/processors/modules/face_swapper.py

from argparse import ArgumentParser
from typing import List, Tuple

import numpy

import facefusion.jobs.job_manager
import facefusion.jobs.job_store
import facefusion.processors.core as processors
from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, process_manager, state_manager, wording
from facefusion.common_helper import get_first
from facefusion.download import conditional_download_hashes, conditional_download_sources
from facefusion.execution import has_execution_provider
from facefusion.face_analyser import get_average_face, get_many_faces, get_one_face
from facefusion.face_helper import paste_back, warp_face_by_face_landmark_5
from facefusion.face_masker import create_occlusion_mask, create_region_mask, create_static_box_mask
from facefusion.face_selector import find_similar_faces, sort_and_filter_faces
from facefusion.face_store import get_reference_faces
from facefusion.filesystem import filter_image_paths, has_image, in_directory, is_image, is_video, resolve_relative_path, same_file_extension
from facefusion.inference_manager import get_static_model_initializer
from facefusion.processors import choices as processors_choices
from facefusion.processors.pixel_boost import explode_pixel_boost, implode_pixel_boost
from facefusion.processors.typing import FaceSwapperInputs
from facefusion.program_helper import find_argument_group, suggest_face_swapper_pixel_boost_choices
from facefusion.thread_helper import conditional_thread_semaphore
from facefusion.typing import ApplyStateItem, Args, Embedding, Face, InferencePool, ModelOptions, ModelSet, ProcessMode, QueuePayload, UpdateProgress, VisionFrame
from facefusion.vision import read_image, read_static_image, read_static_images, unpack_resolution, write_image

MODEL_SET : ModelSet =\
{
	'blendswap_256':
	{
		'hashes':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/blendswap_256.hash',
				'path': resolve_relative_path('../.assets/models/blendswap_256.hash')
			}
		},
		'sources':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/blendswap_256.onnx',
				'path': resolve_relative_path('../.assets/models/blendswap_256.onnx')
			}
		},
		'type': 'blendswap',
		'template': 'ffhq_512',
		'size': (256, 256),
		'mean': [ 0.0, 0.0, 0.0 ],
		'standard_deviation': [ 1.0, 1.0, 1.0 ]
	},
	'ghost_1_256':
	{
		'hashes':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/ghost_1_256.hash',
				'path': resolve_relative_path('../.assets/models/ghost_1_256.hash')
			},
			'embedding_converter':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/arcface_converter_ghost.hash',
				'path': resolve_relative_path('../.assets/models/arcface_converter_ghost.hash')
			}
		},
		'sources':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/ghost_1_256.onnx',
				'path': resolve_relative_path('../.assets/models/ghost_1_256.onnx')
			},
			'embedding_converter':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/arcface_converter_ghost.onnx',
				'path': resolve_relative_path('../.assets/models/arcface_converter_ghost.onnx')
			}
		},
		'type': 'ghost',
		'template': 'arcface_112_v1',
		'size': (256, 256),
		'mean': [ 0.5, 0.5, 0.5 ],
		'standard_deviation': [ 0.5, 0.5, 0.5 ]
	},
	'ghost_2_256':
	{
		'hashes':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/ghost_2_256.hash',
				'path': resolve_relative_path('../.assets/models/ghost_2_256.hash')
			},
			'embedding_converter':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/arcface_converter_ghost.hash',
				'path': resolve_relative_path('../.assets/models/arcface_converter_ghost.hash')
			}
		},
		'sources':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/ghost_2_256.onnx',
				'path': resolve_relative_path('../.assets/models/ghost_2_256.onnx')
			},
			'embedding_converter':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/arcface_converter_ghost.onnx',
				'path': resolve_relative_path('../.assets/models/arcface_converter_ghost.onnx')
			}
		},
		'type': 'ghost',
		'template': 'arcface_112_v1',
		'size': (256, 256),
		'mean': [ 0.5, 0.5, 0.5 ],
		'standard_deviation': [ 0.5, 0.5, 0.5 ]
	},
	'ghost_3_256':
	{
		'hashes':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/ghost_3_256.hash',
				'path': resolve_relative_path('../.assets/models/ghost_3_256.hash')
			},
			'embedding_converter':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/arcface_converter_ghost.hash',
				'path': resolve_relative_path('../.assets/models/arcface_converter_ghost.hash')
			}
		},
		'sources':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/ghost_3_256.onnx',
				'path': resolve_relative_path('../.assets/models/ghost_3_256.onnx')
			},
			'embedding_converter':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/arcface_converter_ghost.onnx',
				'path': resolve_relative_path('../.assets/models/arcface_converter_ghost.onnx')
			}
		},
		'type': 'ghost',
		'template': 'arcface_112_v1',
		'size': (256, 256),
		'mean': [ 0.5, 0.5, 0.5 ],
		'standard_deviation': [ 0.5, 0.5, 0.5 ]
	},
	'inswapper_128':
	{
		'hashes':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/inswapper_128.hash',
				'path': resolve_relative_path('../.assets/models/inswapper_128.hash')
			}
		},
		'sources':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/inswapper_128.onnx',
				'path': resolve_relative_path('../.assets/models/inswapper_128.onnx')
			}
		},
		'type': 'inswapper',
		'template': 'arcface_128_v2',
		'size': (128, 128),
		'mean': [ 0.0, 0.0, 0.0 ],
		'standard_deviation': [ 1.0, 1.0, 1.0 ]
	},
	'inswapper_128_fp16':
	{
		'hashes':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/inswapper_128_fp16.hash',
				'path': resolve_relative_path('../.assets/models/inswapper_128_fp16.hash')
			}
		},
		'sources':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/inswapper_128_fp16.onnx',
				'path': resolve_relative_path('../.assets/models/inswapper_128_fp16.onnx')
			}
		},
		'type': 'inswapper',
		'template': 'arcface_128_v2',
		'size': (128, 128),
		'mean': [ 0.0, 0.0, 0.0 ],
		'standard_deviation': [ 1.0, 1.0, 1.0 ]
	},
	'simswap_256':
	{
		'hashes':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/simswap_256.hash',
				'path': resolve_relative_path('../.assets/models/simswap_256.hash')
			},
			'embedding_converter':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/arcface_converter_simswap.hash',
				'path': resolve_relative_path('../.assets/models/arcface_converter_simswap.hash')
			}
		},
		'sources':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/simswap_256.onnx',
				'path': resolve_relative_path('../.assets/models/simswap_256.onnx')
			},
			'embedding_converter':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/arcface_converter_simswap.onnx',
				'path': resolve_relative_path('../.assets/models/arcface_converter_simswap.onnx')
			}
		},
		'type': 'simswap',
		'template': 'arcface_112_v1',
		'size': (256, 256),
		'mean': [ 0.485, 0.456, 0.406 ],
		'standard_deviation': [ 0.229, 0.224, 0.225 ]
	},
	'simswap_unofficial_512':
	{
		'hashes':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/simswap_unofficial_512.hash',
				'path': resolve_relative_path('../.assets/models/simswap_unofficial_512.hash')
			},
			'embedding_converter':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/arcface_converter_simswap.hash',
				'path': resolve_relative_path('../.assets/models/arcface_converter_simswap.hash')
			}
		},
		'sources':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/simswap_unofficial_512.onnx',
				'path': resolve_relative_path('../.assets/models/simswap_unofficial_512.onnx')
			},
			'embedding_converter':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/arcface_converter_simswap.onnx',
				'path': resolve_relative_path('../.assets/models/arcface_converter_simswap.onnx')
			}
		},
		'type': 'simswap',
		'template': 'arcface_112_v1',
		'size': (512, 512),
		'mean': [ 0.0, 0.0, 0.0 ],
		'standard_deviation': [ 1.0, 1.0, 1.0 ]
	},
	'uniface_256':
	{
		'hashes':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/uniface_256.hash',
				'path': resolve_relative_path('../.assets/models/uniface_256.hash')
			}
		},
		'sources':
		{
			'face_swapper':
			{
				'url': 'https://github.com/facefusion/facefusion-assets/releases/download/models-3.0.0/uniface_256.onnx',
				'path': resolve_relative_path('../.assets/models/uniface_256.onnx')
			}
		},
		'type': 'uniface',
		'template': 'ffhq_512',
		'size': (256, 256),
		'mean': [ 0.5, 0.5, 0.5 ],
		'standard_deviation': [ 0.5, 0.5, 0.5 ]
	}
}


def get_inference_pool() -> InferencePool:
	model_sources = get_model_options().get('sources')
	model_context = __name__ + '.' + state_manager.get_item('face_swapper_model')
	return inference_manager.get_inference_pool(model_context, model_sources)


def clear_inference_pool() -> None:
	model_context = __name__ + '.' + state_manager.get_item('face_swapper_model')
	inference_manager.clear_inference_pool(model_context)


def get_model_options() -> ModelOptions:
	face_swapper_model = state_manager.get_item('face_swapper_model')
	face_swapper_model = 'inswapper_128' if has_execution_provider('coreml') and face_swapper_model == 'inswapper_128_fp16' else face_swapper_model
	return MODEL_SET.get(face_swapper_model)


def register_args(program : ArgumentParser) -> None:
	group_processors = find_argument_group(program, 'processors')
	if group_processors:
		group_processors.add_argument('--face-swapper-model', help = wording.get('help.face_swapper_model'), default = config.get_str_value('processors.face_swapper_model', 'inswapper_128_fp16'), choices = processors_choices.face_swapper_set.keys())
		face_swapper_pixel_boost_choices = suggest_face_swapper_pixel_boost_choices(program)
		group_processors.add_argument('--face-swapper-pixel-boost', help = wording.get('help.face_swapper_pixel_boost'), default = config.get_str_value('processors.face_swapper_pixel_boost', get_first(face_swapper_pixel_boost_choices)), choices = face_swapper_pixel_boost_choices)
		facefusion.jobs.job_store.register_step_keys([ 'face_swapper_model', 'face_swapper_pixel_boost' ])


def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
	apply_state_item('face_swapper_model', args.get('face_swapper_model'))
	apply_state_item('face_swapper_pixel_boost', args.get('face_swapper_pixel_boost'))


def pre_check() -> bool:
	download_directory_path = resolve_relative_path('../.assets/models')
	model_hashes = get_model_options().get('hashes')
	model_sources = get_model_options().get('sources')

	return conditional_download_hashes(download_directory_path, model_hashes) and conditional_download_sources(download_directory_path, model_sources)


def pre_process(mode : ProcessMode) -> bool:
	if not has_image(state_manager.get_item('source_paths')):
		logger.error(wording.get('choose_image_source') + wording.get('exclamation_mark'), __name__)
		return False
	source_image_paths = filter_image_paths(state_manager.get_item('source_paths'))
	source_frames = read_static_images(source_image_paths)
	source_faces = get_many_faces(source_frames)
	if not get_one_face(source_faces):
		logger.error(wording.get('no_source_face_detected') + wording.get('exclamation_mark'), __name__)
		return False
	if mode in [ 'output', 'preview' ] and not is_image(state_manager.get_item('target_path')) and not is_video(state_manager.get_item('target_path')):
		logger.error(wording.get('choose_image_or_video_target') + wording.get('exclamation_mark'), __name__)
		return False
	if mode == 'output' and not in_directory(state_manager.get_item('output_path')):
		logger.error(wording.get('specify_image_or_video_output') + wording.get('exclamation_mark'), __name__)
		return False
	if mode == 'output' and not same_file_extension([ state_manager.get_item('target_path'), state_manager.get_item('output_path') ]):
		logger.error(wording.get('match_target_and_output_extension') + wording.get('exclamation_mark'), __name__)
		return False
	return True


def post_process() -> None:
	read_static_image.cache_clear()
	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
		clear_inference_pool()
		get_static_model_initializer.cache_clear()
	if state_manager.get_item('video_memory_strategy') == 'strict':
		content_analyser.clear_inference_pool()
		face_classifier.clear_inference_pool()
		face_detector.clear_inference_pool()
		face_landmarker.clear_inference_pool()
		face_masker.clear_inference_pool()
		face_recognizer.clear_inference_pool()


def swap_face(source_face : Face, target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
	model_template = get_model_options().get('template')
	model_size = get_model_options().get('size')
	pixel_boost_size = unpack_resolution(state_manager.get_item('face_swapper_pixel_boost'))
	pixel_boost_total = pixel_boost_size[0] // model_size[0]
	crop_vision_frame, affine_matrix = warp_face_by_face_landmark_5(temp_vision_frame, target_face.landmark_set.get('5/68'), model_template, pixel_boost_size)
	temp_vision_frames = []
	crop_masks = []

	if 'box' in state_manager.get_item('face_mask_types'):
		box_mask = create_static_box_mask(crop_vision_frame.shape[:2][::-1], state_manager.get_item('face_mask_blur'), state_manager.get_item('face_mask_padding'))
		crop_masks.append(box_mask)

	if 'occlusion' in state_manager.get_item('face_mask_types'):
		occlusion_mask = create_occlusion_mask(crop_vision_frame)
		crop_masks.append(occlusion_mask)

	pixel_boost_vision_frames = implode_pixel_boost(crop_vision_frame, pixel_boost_total, model_size)
	for pixel_boost_vision_frame in pixel_boost_vision_frames:
		pixel_boost_vision_frame = prepare_crop_frame(pixel_boost_vision_frame)
		pixel_boost_vision_frame = forward_swap_face(source_face, pixel_boost_vision_frame)
		pixel_boost_vision_frame = normalize_crop_frame(pixel_boost_vision_frame)
		temp_vision_frames.append(pixel_boost_vision_frame)
	crop_vision_frame = explode_pixel_boost(temp_vision_frames, pixel_boost_total, model_size, pixel_boost_size)

	if 'region' in state_manager.get_item('face_mask_types'):
		region_mask = create_region_mask(crop_vision_frame, state_manager.get_item('face_mask_regions'))
		crop_masks.append(region_mask)

	crop_mask = numpy.minimum.reduce(crop_masks).clip(0, 1)
	temp_vision_frame = paste_back(temp_vision_frame, crop_vision_frame, crop_mask, affine_matrix)
	return temp_vision_frame


def forward_swap_face(source_face : Face, crop_vision_frame : VisionFrame) -> VisionFrame:
	face_swapper = get_inference_pool().get('face_swapper')
	model_type = get_model_options().get('type')
	face_swapper_inputs = {}

	for face_swapper_input in face_swapper.get_inputs():
		if face_swapper_input.name == 'source':
			if model_type == 'blendswap' or model_type == 'uniface':
				face_swapper_inputs[face_swapper_input.name] = prepare_source_frame(source_face)
			else:
				face_swapper_inputs[face_swapper_input.name] = prepare_source_embedding(source_face)
		if face_swapper_input.name == 'target':
			face_swapper_inputs[face_swapper_input.name] = crop_vision_frame

	with conditional_thread_semaphore():
		crop_vision_frame = face_swapper.run(None, face_swapper_inputs)[0][0]

	return crop_vision_frame


def forward_convert_embedding(embedding : Embedding) -> Embedding:
	embedding_converter = get_inference_pool().get('embedding_converter')

	with conditional_thread_semaphore():
		embedding = embedding_converter.run(None,
		{
			'input': embedding
		})[0]

	return embedding


def prepare_source_frame(source_face : Face) -> VisionFrame:
	model_type = get_model_options().get('type')
	source_vision_frame = read_static_image(get_first(state_manager.get_item('source_paths')))

	if model_type == 'blendswap':
		source_vision_frame, _ = warp_face_by_face_landmark_5(source_vision_frame, source_face.landmark_set.get('5/68'), 'arcface_112_v2', (112, 112))
	if model_type == 'uniface':
		source_vision_frame, _ = warp_face_by_face_landmark_5(source_vision_frame, source_face.landmark_set.get('5/68'), 'ffhq_512', (256, 256))
	source_vision_frame = source_vision_frame[:, :, ::-1] / 255.0
	source_vision_frame = source_vision_frame.transpose(2, 0, 1)
	source_vision_frame = numpy.expand_dims(source_vision_frame, axis = 0).astype(numpy.float32)
	return source_vision_frame


def prepare_source_embedding(source_face : Face) -> Embedding:
	model_type = get_model_options().get('type')

	if model_type == 'ghost':
		source_embedding, _ = convert_embedding(source_face)
		source_embedding = source_embedding.reshape(1, -1)
	elif model_type == 'inswapper':
		model_path = get_model_options().get('sources').get('face_swapper').get('path')
		model_initializer = get_static_model_initializer(model_path)
		source_embedding = source_face.embedding.reshape((1, -1))
		source_embedding = numpy.dot(source_embedding, model_initializer) / numpy.linalg.norm(source_embedding)
	else:
		_, source_normed_embedding = convert_embedding(source_face)
		source_embedding = source_normed_embedding.reshape(1, -1)
	return source_embedding


def convert_embedding(source_face : Face) -> Tuple[Embedding, Embedding]:
	embedding = source_face.embedding.reshape(-1, 512)
	embedding = forward_convert_embedding(embedding)
	embedding = embedding.ravel()
	normed_embedding = embedding / numpy.linalg.norm(embedding)
	return embedding, normed_embedding


def prepare_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:
	model_mean = get_model_options().get('mean')
	model_standard_deviation = get_model_options().get('standard_deviation')

	crop_vision_frame = crop_vision_frame[:, :, ::-1] / 255.0
	crop_vision_frame = (crop_vision_frame - model_mean) / model_standard_deviation
	crop_vision_frame = crop_vision_frame.transpose(2, 0, 1)
	crop_vision_frame = numpy.expand_dims(crop_vision_frame, axis = 0).astype(numpy.float32)
	return crop_vision_frame


def normalize_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:
	model_type = get_model_options().get('type')
	model_mean = get_model_options().get('mean')
	model_standard_deviation = get_model_options().get('standard_deviation')

	crop_vision_frame = crop_vision_frame.transpose(1, 2, 0)
	if model_type == 'ghost' or model_type == 'uniface':
		crop_vision_frame = crop_vision_frame * model_standard_deviation + model_mean
	crop_vision_frame = crop_vision_frame.clip(0, 1)
	crop_vision_frame = crop_vision_frame[:, :, ::-1] * 255
	return crop_vision_frame


def get_reference_frame(source_face : Face, target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
	return swap_face(source_face, target_face, temp_vision_frame)


def process_frame(inputs : FaceSwapperInputs) -> VisionFrame:
	reference_faces = inputs.get('reference_faces')
	source_face = inputs.get('source_face')
	target_vision_frame = inputs.get('target_vision_frame')
	many_faces = sort_and_filter_faces(get_many_faces([ target_vision_frame ]))

	if state_manager.get_item('face_selector_mode') == 'many':
		if many_faces:
			for target_face in many_faces:
				target_vision_frame = swap_face(source_face, target_face, target_vision_frame)
	if state_manager.get_item('face_selector_mode') == 'one':
		target_face = get_one_face(many_faces)
		if target_face:
			target_vision_frame = swap_face(source_face, target_face, target_vision_frame)
	if state_manager.get_item('face_selector_mode') == 'reference':
		similar_faces = find_similar_faces(many_faces, reference_faces, state_manager.get_item('reference_face_distance'))
		if similar_faces:
			for similar_face in similar_faces:
				target_vision_frame = swap_face(source_face, similar_face, target_vision_frame)
	return target_vision_frame


def process_frames(source_paths : List[str], queue_payloads : List[QueuePayload], update_progress : UpdateProgress) -> None:
	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
	source_frames = read_static_images(source_paths)
	source_faces = get_many_faces(source_frames)
	source_face = get_average_face(source_faces)

	for queue_payload in process_manager.manage(queue_payloads):
		target_vision_path = queue_payload['frame_path']
		target_vision_frame = read_image(target_vision_path)
		output_vision_frame = process_frame(
		{
			'reference_faces': reference_faces,
			'source_face': source_face,
			'target_vision_frame': target_vision_frame
		})
		write_image(target_vision_path, output_vision_frame)
		update_progress(1)


def process_image(source_paths : List[str], target_path : str, output_path : str) -> None:
	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
	source_frames = read_static_images(source_paths)
	source_faces = get_many_faces(source_frames)
	source_face = get_average_face(source_faces)
	target_vision_frame = read_static_image(target_path)
	output_vision_frame = process_frame(
	{
		'reference_faces': reference_faces,
		'source_face': source_face,
		'target_vision_frame': target_vision_frame
	})
	write_image(output_path, output_vision_frame)


def process_video(source_paths : List[str], temp_frame_paths : List[str]) -> None:
	processors.multi_process_frames(source_paths, temp_frame_paths, process_frames)