mirror of
https://github.com/s0md3v/roop.git
synced 2025-09-26 20:31:16 +08:00
fe9b2bc8e5
* Fix video frames lower than threads * Skip target audio (#656) * Improve return typing * Use face enhancer device according to execution provider * Lock face by reference (#679) * Lock face by position of a face reference * Prevent exception for get_many_faces * Finalize face reference implementation * Fix potential exception * Use sys.exit() over quit() * Split frame processor error to reduce confusion * Improve face reference by introducing more CLI args * Prevent AttributeError if face is None * Update dependencies * Move reference creation to process_video * Allow to initialize UI with source path and target path * Allow to initialize UI with source path and target path * Allow to initialize UI with source path and target path * Use onnxruntime-coreml for old MacOS * Fix typing * Fix typing * Fix typing * Temp fix for enhancer * Temp fix for enhancer * Keyboard bindings to change reference face via Up/Down * Fix slow preview * ignore * Update README and ISSUE TEMPLATES * Right/Left to update frames by +10/-10 * Fix fps mismatch * Add fps parameter to extract_frames() * Minor wording cosmetics * Improve enhancer performance by using cropped face * Fix suggested threads and memory * Extract frames with FPS output * Remove default max-memory * Remove release_resources() as it does not work * Ignore torch import * Add drag and drop for source and target * Fix typing * Bump version * Limit Left/Right binding to videos * Add key binding hits to preview
97 lines
3.0 KiB
Python
97 lines
3.0 KiB
Python
from typing import Any, List, Callable
|
|
import cv2
|
|
import threading
|
|
from gfpgan.utils import GFPGANer
|
|
|
|
import roop.globals
|
|
import roop.processors.frame.core
|
|
from roop.core import update_status
|
|
from roop.face_analyser import get_many_faces
|
|
from roop.typing import Frame, Face
|
|
from roop.utilities import conditional_download, resolve_relative_path, is_image, is_video
|
|
|
|
FACE_ENHANCER = None
|
|
THREAD_SEMAPHORE = threading.Semaphore()
|
|
THREAD_LOCK = threading.Lock()
|
|
NAME = 'ROOP.FACE-ENHANCER'
|
|
|
|
|
|
def get_face_enhancer() -> Any:
|
|
global FACE_ENHANCER
|
|
|
|
with THREAD_LOCK:
|
|
if FACE_ENHANCER is None:
|
|
model_path = resolve_relative_path('../models/GFPGANv1.4.pth')
|
|
# todo: set models path -> https://github.com/TencentARC/GFPGAN/issues/399
|
|
FACE_ENHANCER = GFPGANer(model_path=model_path, upscale=1, device=get_device())
|
|
return FACE_ENHANCER
|
|
|
|
|
|
def get_device() -> str:
|
|
if 'CUDAExecutionProvider' in roop.globals.execution_providers:
|
|
return 'cuda'
|
|
if 'CoreMLExecutionProvider' in roop.globals.execution_providers:
|
|
return 'mps'
|
|
return 'cpu'
|
|
|
|
|
|
def clear_face_enhancer() -> None:
|
|
global FACE_ENHANCER
|
|
|
|
FACE_ENHANCER = None
|
|
|
|
|
|
def pre_check() -> bool:
|
|
download_directory_path = resolve_relative_path('../models')
|
|
conditional_download(download_directory_path, ['https://huggingface.co/henryruhs/roop/resolve/main/GFPGANv1.4.pth'])
|
|
return True
|
|
|
|
|
|
def pre_start() -> bool:
|
|
if not is_image(roop.globals.target_path) and not is_video(roop.globals.target_path):
|
|
update_status('Select an image or video for target path.', NAME)
|
|
return False
|
|
return True
|
|
|
|
|
|
def post_process() -> None:
|
|
clear_face_enhancer()
|
|
|
|
|
|
def enhance_face(target_face: Face, temp_frame: Frame) -> Frame:
|
|
start_x, start_y, end_x, end_y = map(int, target_face['bbox'])
|
|
with THREAD_SEMAPHORE:
|
|
_, _, temp_face = get_face_enhancer().enhance(
|
|
temp_frame[start_y:end_y, start_x:end_x],
|
|
paste_back=True
|
|
)
|
|
temp_frame[start_y:end_y, start_x:end_x] = temp_face
|
|
return temp_frame
|
|
|
|
|
|
def process_frame(source_face: Face, reference_face: Face, temp_frame: Frame) -> Frame:
|
|
many_faces = get_many_faces(temp_frame)
|
|
if many_faces:
|
|
for target_face in many_faces:
|
|
temp_frame = enhance_face(target_face, temp_frame)
|
|
return temp_frame
|
|
|
|
|
|
def process_frames(source_path: str, temp_frame_paths: List[str], update: Callable[[], None]) -> None:
|
|
for temp_frame_path in temp_frame_paths:
|
|
temp_frame = cv2.imread(temp_frame_path)
|
|
result = process_frame(None, None, temp_frame)
|
|
cv2.imwrite(temp_frame_path, result)
|
|
if update:
|
|
update()
|
|
|
|
|
|
def process_image(source_path: str, target_path: str, output_path: str) -> None:
|
|
target_frame = cv2.imread(target_path)
|
|
result = process_frame(None, None, target_frame)
|
|
cv2.imwrite(output_path, result)
|
|
|
|
|
|
def process_video(source_path: str, temp_frame_paths: List[str]) -> None:
|
|
roop.processors.frame.core.process_video(None, temp_frame_paths, process_frames)
|