processors/modules/frame_enhancer/core.py

from argparse import ArgumentParser
from functools import lru_cache
from types import ModuleType
from typing import List

import cv2
import numpy

import facefusion.choices
import facefusion.jobs.job_manager
import facefusion.jobs.job_store
from facefusion import config, content_analyser, inference_manager, logger, state_manager, translator, video_manager
from facefusion.common_helper import create_int_metavar, is_macos
from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
from facefusion.execution import has_execution_provider
from facefusion.filesystem import in_directory, is_image, is_video, resolve_relative_path, same_file_extension
from facefusion.processors.modules.frame_enhancer import choices as frame_enhancer_choices
from facefusion.processors.modules.frame_enhancer.types import FrameEnhancerInputs
from facefusion.processors.types import ProcessorOutputs
from facefusion.program_helper import find_argument_group
from facefusion.thread_helper import conditional_thread_semaphore
from facefusion.types import ApplyStateItem, Args, DownloadScope, InferencePool, InferenceProvider, ModelOptions, ModelSet, ProcessMode, VisionFrame
from facefusion.vision import blend_frame, create_tile_frames, merge_tile_frames, read_static_image, read_static_video_frame


@lru_cache()
def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
	return\
	{
		'clear_reality_x4':
		{
			'__metadata__':
			{
				'vendor': 'Kim2091',
				'license': 'Non-Commercial',
				'year': 2023
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'clear_reality_x4.hash'),
					'path': resolve_relative_path('../.assets/models/clear_reality_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'clear_reality_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/clear_reality_x4.onnx')
				}
			},
			'size': (128, 8, 4),
			'scale': 4
		},
		'face_dat_x4':
		{
			'__metadata__':
			{
				'vendor': 'Helaman',
				'license': 'CC-BY-4.0',
				'year': 2023
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.5.0', 'face_dat_x4.hash'),
					'path': resolve_relative_path('../.assets/models/face_dat_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.5.0', 'face_dat_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/face_dat_x4.onnx')
				}
			},
			'size': (128, 8, 4),
			'scale': 4
		},
		'nomos8k_sc_x4':
		{
			'__metadata__':
			{
				'vendor': 'Phhofm',
				'license': 'CC-BY-4.0',
				'year': 2023
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'nomos8k_sc_x4.hash'),
					'path': resolve_relative_path('../.assets/models/nomos8k_sc_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'nomos8k_sc_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/nomos8k_sc_x4.onnx')
				}
			},
			'size': (128, 8, 4),
			'scale': 4
		},
		'real_esrgan_x2':
		{
			'__metadata__':
			{
				'vendor': 'xinntao',
				'license': 'BSD-3-Clause',
				'year': 2021
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_esrgan_x2.hash'),
					'path': resolve_relative_path('../.assets/models/real_esrgan_x2.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_esrgan_x2.onnx'),
					'path': resolve_relative_path('../.assets/models/real_esrgan_x2.onnx')
				}
			},
			'size': (256, 16, 8),
			'scale': 2
		},
		'real_esrgan_x2_fp16':
		{
			'__metadata__':
			{
				'vendor': 'xinntao',
				'license': 'BSD-3-Clause',
				'year': 2021
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_esrgan_x2_fp16.hash'),
					'path': resolve_relative_path('../.assets/models/real_esrgan_x2_fp16.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_esrgan_x2_fp16.onnx'),
					'path': resolve_relative_path('../.assets/models/real_esrgan_x2_fp16.onnx')
				}
			},
			'precision': 'fp16',
			'size': (256, 16, 8),
			'scale': 2
		},
		'real_esrgan_x4':
		{
			'__metadata__':
			{
				'vendor': 'xinntao',
				'license': 'BSD-3-Clause',
				'year': 2021
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_esrgan_x4.hash'),
					'path': resolve_relative_path('../.assets/models/real_esrgan_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_esrgan_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/real_esrgan_x4.onnx')
				}
			},
			'size': (256, 16, 8),
			'scale': 4
		},
		'real_esrgan_x4_fp16':
		{
			'__metadata__':
			{
				'vendor': 'xinntao',
				'license': 'BSD-3-Clause',
				'year': 2021
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_esrgan_x4_fp16.hash'),
					'path': resolve_relative_path('../.assets/models/real_esrgan_x4_fp16.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_esrgan_x4_fp16.onnx'),
					'path': resolve_relative_path('../.assets/models/real_esrgan_x4_fp16.onnx')
				}
			},
			'precision': 'fp16',
			'size': (256, 16, 8),
			'scale': 4
		},
		'real_esrgan_x8':
		{
			'__metadata__':
			{
				'vendor': 'xinntao',
				'license': 'BSD-3-Clause',
				'year': 2021
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_esrgan_x8.hash'),
					'path': resolve_relative_path('../.assets/models/real_esrgan_x8.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_esrgan_x8.onnx'),
					'path': resolve_relative_path('../.assets/models/real_esrgan_x8.onnx')
				}
			},
			'size': (256, 16, 8),
			'scale': 8
		},
		'real_esrgan_x8_fp16':
		{
			'__metadata__':
			{
				'vendor': 'xinntao',
				'license': 'BSD-3-Clause',
				'year': 2021
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_esrgan_x8_fp16.hash'),
					'path': resolve_relative_path('../.assets/models/real_esrgan_x8_fp16.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_esrgan_x8_fp16.onnx'),
					'path': resolve_relative_path('../.assets/models/real_esrgan_x8_fp16.onnx')
				}
			},
			'precision': 'fp16',
			'size': (256, 16, 8),
			'scale': 8
		},
		'real_hatgan_x4':
		{
			'__metadata__':
			{
				'vendor': 'XPixelGroup',
				'license': 'Apache-2.0',
				'year': 2023
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_hatgan_x4.hash'),
					'path': resolve_relative_path('../.assets/models/real_hatgan_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'real_hatgan_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/real_hatgan_x4.onnx')
				}
			},
			'size': (256, 16, 8),
			'scale': 4
		},
		'real_web_photo_x4':
		{
			'__metadata__':
			{
				'vendor': 'Helaman',
				'license': 'CC-BY-4.0',
				'year': 2024
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.1.0', 'real_web_photo_x4.hash'),
					'path': resolve_relative_path('../.assets/models/real_web_photo_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.1.0', 'real_web_photo_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/real_web_photo_x4.onnx')
				}
			},
			'size': (64, 4, 2),
			'scale': 4
		},
		'realistic_rescaler_x4':
		{
			'__metadata__':
			{
				'vendor': 'Mutin Choler',
				'license': 'WTFPL',
				'year': 2023
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.1.0', 'realistic_rescaler_x4.hash'),
					'path': resolve_relative_path('../.assets/models/realistic_rescaler_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.1.0', 'realistic_rescaler_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/realistic_rescaler_x4.onnx')
				}
			},
			'size': (128, 8, 4),
			'scale': 4
		},
		'remacri_x4':
		{
			'__metadata__':
			{
				'vendor': 'FoolhardyVEVO',
				'license': 'Non-Commercial',
				'year': 2021
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.1.0', 'remacri_x4.hash'),
					'path': resolve_relative_path('../.assets/models/remacri_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.1.0', 'remacri_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/remacri_x4.onnx')
				}
			},
			'size': (128, 8, 4),
			'scale': 4
		},
		'siax_x4':
		{
			'__metadata__':
			{
				'vendor': 'NMKD',
				'license': 'WTFPL',
				'year': 2021
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.1.0', 'siax_x4.hash'),
					'path': resolve_relative_path('../.assets/models/siax_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.1.0', 'siax_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/siax_x4.onnx')
				}
			},
			'size': (128, 8, 4),
			'scale': 4
		},
		'span_kendata_x4':
		{
			'__metadata__':
			{
				'vendor': 'terrainer',
				'license': 'Non-Commercial',
				'year': 2024
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'span_kendata_x4.hash'),
					'path': resolve_relative_path('../.assets/models/span_kendata_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'span_kendata_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/span_kendata_x4.onnx')
				}
			},
			'size': (128, 8, 4),
			'scale': 4
		},
		'swin2_sr_x4':
		{
			'__metadata__':
			{
				'vendor': 'mv-lab',
				'license': 'Apache-2.0',
				'year': 2022
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.1.0', 'swin2_sr_x4.hash'),
					'path': resolve_relative_path('../.assets/models/swin2_sr_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.1.0', 'swin2_sr_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/swin2_sr_x4.onnx')
				}
			},
			'size': (128, 8, 4),
			'scale': 4
		},
		'tghq_face_x8':
		{
			'__metadata__':
			{
				'vendor': 'TorrentGuy',
				'license': 'GPL-3.0',
				'year': 2019
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.5.0', 'tghq_face_x8.hash'),
					'path': resolve_relative_path('../.assets/models/tghq_face_x8.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.5.0', 'tghq_face_x8.onnx'),
					'path': resolve_relative_path('../.assets/models/tghq_face_x8.onnx')
				}
			},
			'size': (128, 8, 4),
			'scale': 8
		},
		'ultra_sharp_x4':
		{
			'__metadata__':
			{
				'vendor': 'Kim2091',
				'license': 'Non-Commercial',
				'year': 2021
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'ultra_sharp_x4.hash'),
					'path': resolve_relative_path('../.assets/models/ultra_sharp_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.0.0', 'ultra_sharp_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/ultra_sharp_x4.onnx')
				}
			},
			'size': (128, 8, 4),
			'scale': 4
		},
		'ultra_sharp_2_x4':
		{
			'__metadata__':
			{
				'vendor': 'Kim2091',
				'license': 'Non-Commercial',
				'year': 2025
			},
			'hashes':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.3.0', 'ultra_sharp_2_x4.hash'),
					'path': resolve_relative_path('../.assets/models/ultra_sharp_2_x4.hash')
				}
			},
			'sources':
			{
				'frame_enhancer':
				{
					'url': resolve_download_url('models-3.3.0', 'ultra_sharp_2_x4.onnx'),
					'path': resolve_relative_path('../.assets/models/ultra_sharp_2_x4.onnx')
				}
			},
			'size': (1024, 64, 32),
			'scale': 4
		}
	}


def get_inference_pool() -> InferencePool:
	model_names = [ state_manager.get_item('frame_enhancer_model') ]
	model_source_set = get_model_options().get('sources')

	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)


def clear_inference_pool() -> None:
	model_names = [ state_manager.get_item('frame_enhancer_model') ]
	inference_manager.clear_inference_pool(__name__, model_names)


def adjust_inference_providers() -> List[InferenceProvider]:
	model_precision = get_model_options().get('precision')

	if is_macos() and has_execution_provider('coreml') and model_precision == 'fp16':
		return\
		[
			(facefusion.choices.execution_provider_set.get('coreml'),
			{
				'ModelFormat': 'MLProgram'
			})
		]

	return []


def get_model_options() -> ModelOptions:
	model_name = state_manager.get_item('frame_enhancer_model')
	return create_static_model_set('full').get(model_name)


def register_args(program : ArgumentParser) -> None:
	group_processors = find_argument_group(program, 'processors')
	if group_processors:
		group_processors.add_argument('--frame-enhancer-model', help = translator.get('help.model', __package__), default = config.get_str_value('processors', 'frame_enhancer_model', 'span_kendata_x4'), choices = frame_enhancer_choices.frame_enhancer_models)
		group_processors.add_argument('--frame-enhancer-blend', help = translator.get('help.blend', __package__), type = int, default = config.get_int_value('processors', 'frame_enhancer_blend', '80'), choices = frame_enhancer_choices.frame_enhancer_blend_range, metavar = create_int_metavar(frame_enhancer_choices.frame_enhancer_blend_range))
		facefusion.jobs.job_store.register_step_keys([ 'frame_enhancer_model', 'frame_enhancer_blend' ])


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


def get_common_modules() -> List[ModuleType]:
	return [ content_analyser ]


def pre_check() -> bool:
	model_hash_set = get_model_options().get('hashes')
	model_source_set = get_model_options().get('sources')

	for common_module in get_common_modules():
		if not common_module.pre_check():
			return False

	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)


def pre_process(mode : ProcessMode) -> bool:
	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(translator.get('choose_image_or_video_target') + translator.get('exclamation_mark'), __name__)
		return False
	if mode == 'output' and not in_directory(state_manager.get_item('output_path')):
		logger.error(translator.get('specify_image_or_video_output') + translator.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(translator.get('match_target_and_output_extension') + translator.get('exclamation_mark'), __name__)
		return False
	return True


def post_process() -> None:
	read_static_image.cache_clear()
	read_static_video_frame.cache_clear()
	video_manager.clear_video_pool()

	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
		clear_inference_pool()

	if state_manager.get_item('video_memory_strategy') == 'strict':
		for common_module in get_common_modules():
			common_module.clear_inference_pool()


def enhance_frame(temp_vision_frame : VisionFrame) -> VisionFrame:
	model_size = get_model_options().get('size')
	model_scale = get_model_options().get('scale')
	temp_height, temp_width = temp_vision_frame.shape[:2]
	tile_vision_frames, pad_width, pad_height = create_tile_frames(temp_vision_frame, model_size)

	for index, tile_vision_frame in enumerate(tile_vision_frames):
		tile_vision_frame = prepare_tile_frame(tile_vision_frame)
		tile_vision_frame = forward(tile_vision_frame)
		tile_vision_frames[index] = normalize_tile_frame(tile_vision_frame)

	merge_vision_frame = merge_tile_frames(tile_vision_frames, temp_width * model_scale, temp_height * model_scale, pad_width * model_scale, pad_height * model_scale, (model_size[0] * model_scale, model_size[1] * model_scale, model_size[2] * model_scale))
	temp_vision_frame = blend_merge_frame(temp_vision_frame, merge_vision_frame)
	return temp_vision_frame


def forward(tile_vision_frame : VisionFrame) -> VisionFrame:
	frame_enhancer = get_inference_pool().get('frame_enhancer')

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

	return tile_vision_frame


def prepare_tile_frame(tile_vision_frame : VisionFrame) -> VisionFrame:
	tile_vision_frame = numpy.expand_dims(tile_vision_frame[:, :, ::-1], axis = 0)
	tile_vision_frame = tile_vision_frame.transpose(0, 3, 1, 2)
	tile_vision_frame = tile_vision_frame.astype(numpy.float32) / 255.0
	return tile_vision_frame


def normalize_tile_frame(tile_vision_frame : VisionFrame) -> VisionFrame:
	tile_vision_frame = tile_vision_frame.transpose(0, 2, 3, 1).squeeze(0) * 255
	tile_vision_frame = tile_vision_frame.clip(0, 255).astype(numpy.uint8)[:, :, ::-1]
	return tile_vision_frame


def blend_merge_frame(temp_vision_frame : VisionFrame, merge_vision_frame : VisionFrame) -> VisionFrame:
	frame_enhancer_blend = 1 - (state_manager.get_item('frame_enhancer_blend') / 100)
	temp_vision_frame = cv2.resize(temp_vision_frame, (merge_vision_frame.shape[1], merge_vision_frame.shape[0]))
	temp_vision_frame = blend_frame(temp_vision_frame, merge_vision_frame, 1 - frame_enhancer_blend)
	return temp_vision_frame


def process_frame(inputs : FrameEnhancerInputs) -> ProcessorOutputs:
	temp_vision_frame = inputs.get('temp_vision_frame')
	temp_vision_mask = inputs.get('temp_vision_mask')
	temp_vision_frame = enhance_frame(temp_vision_frame)
	temp_vision_mask = cv2.resize(temp_vision_mask, temp_vision_frame.shape[:2][::-1])
	return temp_vision_frame, temp_vision_mask