Files
facefusion/facefusion/vision.py
Henry Ruhs 3a5fe2a602 2.1.0 (#253)
* Operating system specific installer options

* Update dependencies

* Sorting before NMS according to the standard

* Minor typing fix

* Change the wording

* Update preview.py (#222)

Added a release listener to the preview frame slider, this will update the frame preview with the latest frame

* Combine preview slider listener

* Remove change listener

* Introduce multi source (#223)

* Implement multi source

* Adjust face enhancer and face debugger to multi source

* Implement multi source to UI

* Implement multi source to UI part2

* Implement multi source to UI part3

* Implement multi source to UI part4

* Some cleanup

* Add face occluder (#225) (#226)

* Add face occluder (#225)

* add face-occluder (commandline only)

* review 1

* Update face_masker.py

* Update face_masker.py

* Add gui & fix typing

* Minor naming cleanup

* Minor naming cleanup part2

---------

Co-authored-by: Harisreedhar <46858047+harisreedhar@users.noreply.github.com>

* Update usage information

* Fix averaged normed_embedding

* Remove blur from face occluder, enable accelerators

* Switch to RANSAC with 100 threshold

* Update face_enhancer.py (#229)

* Update face_debugger.py (#230)

* Split utilities (#232)

* Split utilities

* Split utilities part2

* Split utilities part3

* Split utilities part4

* Some cleanup

* Implement log level support (#233)

* Implement log level support

* Fix testing

* Implement debug logger

* Implement debug logger

* Fix alignment offset (#235)

* Update face_helper.py

* fix 2

* Enforce virtual environment via installer

* Enforce virtual environment via installer

* Enforce virtual environment via installer

* Enforce virtual environment via installer

* Feat/multi process reference faces (#239)

* Multi processing aware reference faces

* First clean up and joining of files

* Finalize the face store

* Reduce similar face detection to one set, use __name__ for scopes in logger

* Rename to face_occluder

* Introduce ModelSet type

* Improve webcam error handling

* Prevent null pointer on is_image() and is_video()

* Prevent null pointer on is_image() and is_video()

* Fix find similar faces

* Fix find similar faces

* Fix process_images for face enhancer

* Bunch of minor improvements

* onnxruntime for ROCM under linux

* Improve mask related naming

* Fix falsy import

* Fix typo

* Feat/face parser refactoring (#247)

* Face parser update (#244)

* face-parser

* Update face_masker.py

* update debugger

* Update globals.py

* Update face_masker.py

* Refactor code to split occlusion from region

* fix (#246)

* fix

* fix debugger resolution

* flip input to horizontal

* Clean up UI

* Reduce the regions to inside face only

* Reduce the regions to inside face only

---------

Co-authored-by: Harisreedhar <46858047+harisreedhar@users.noreply.github.com>

* Fix enhancer, remove useless dest in add_argument()

* Prevent unselect of the face_mask_regions via UI

* Prepare next release

* Shorten arguments that have choices and nargs

* Add missing clear to face debugger

---------

Co-authored-by: Mathias <github@feroc.de>
Co-authored-by: Harisreedhar <46858047+harisreedhar@users.noreply.github.com>
2023-12-20 00:00:32 +01:00

76 lines
2.0 KiB
Python

from typing import Optional, List
from functools import lru_cache
import cv2
from facefusion.typing import Frame
def get_video_frame(video_path : str, frame_number : int = 0) -> Optional[Frame]:
if video_path:
video_capture = cv2.VideoCapture(video_path)
if video_capture.isOpened():
frame_total = video_capture.get(cv2.CAP_PROP_FRAME_COUNT)
video_capture.set(cv2.CAP_PROP_POS_FRAMES, min(frame_total, frame_number - 1))
has_frame, frame = video_capture.read()
video_capture.release()
if has_frame:
return frame
return None
def detect_fps(video_path : str) -> Optional[float]:
if video_path:
video_capture = cv2.VideoCapture(video_path)
if video_capture.isOpened():
return video_capture.get(cv2.CAP_PROP_FPS)
return None
def count_video_frame_total(video_path : str) -> int:
if video_path:
video_capture = cv2.VideoCapture(video_path)
if video_capture.isOpened():
video_frame_total = int(video_capture.get(cv2.CAP_PROP_FRAME_COUNT))
video_capture.release()
return video_frame_total
return 0
def normalize_frame_color(frame : Frame) -> Frame:
return cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
def resize_frame_dimension(frame : Frame, max_width : int, max_height : int) -> Frame:
height, width = frame.shape[:2]
if height > max_height or width > max_width:
scale = min(max_height / height, max_width / width)
new_width = int(width * scale)
new_height = int(height * scale)
return cv2.resize(frame, (new_width, new_height))
return frame
@lru_cache(maxsize = 128)
def read_static_image(image_path : str) -> Optional[Frame]:
return read_image(image_path)
def read_static_images(image_paths : List[str]) -> Optional[List[Frame]]:
frames = []
if image_paths:
for image_path in image_paths:
frames.append(read_static_image(image_path))
return frames
def read_image(image_path : str) -> Optional[Frame]:
if image_path:
return cv2.imread(image_path)
return None
def write_image(image_path : str, frame : Frame) -> bool:
if image_path:
return cv2.imwrite(image_path, frame)
return False