mirror of
https://github.com/facefusion/facefusion.git
synced 2026-04-30 05:17:49 +02:00
Follow yolo convention, renaming in face detector (#858)
* Follow yolo convention, renaming in face detector * Follow yolo convention, renaming in face detector
This commit is contained in:
Henry Ruhs
@@ -9,7 +9,7 @@ face_detector_set : FaceDetectorSet =\
|
||||
'many': [ '640x640' ],
|
||||
'retinaface': [ '160x160', '320x320', '480x480', '512x512', '640x640' ],
|
||||
'scrfd': [ '160x160', '320x320', '480x480', '512x512', '640x640' ],
|
||||
'yoloface': [ '640x640' ]
|
||||
'yolo_face': [ '640x640' ]
|
||||
}
|
||||
face_detector_models : List[FaceDetectorModel] = list(face_detector_set.keys())
|
||||
face_landmarker_models : List[FaceLandmarkerModel] = [ 'many', '2dfan4', 'peppa_wutz' ]
|
||||
|
||||
+50
-47
@@ -55,11 +55,11 @@ def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
|
||||
}
|
||||
}
|
||||
},
|
||||
'yoloface':
|
||||
'yolo_face':
|
||||
{
|
||||
'hashes':
|
||||
{
|
||||
'yoloface':
|
||||
'yolo_face':
|
||||
{
|
||||
'url': resolve_download_url('models-3.0.0', 'yoloface_8n.hash'),
|
||||
'path': resolve_relative_path('../.assets/models/yoloface_8n.hash')
|
||||
@@ -67,7 +67,7 @@ def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
|
||||
},
|
||||
'sources':
|
||||
{
|
||||
'yoloface':
|
||||
'yolo_face':
|
||||
{
|
||||
'url': resolve_download_url('models-3.0.0', 'yoloface_8n.onnx'),
|
||||
'path': resolve_relative_path('../.assets/models/yoloface_8n.onnx')
|
||||
@@ -91,7 +91,7 @@ def collect_model_downloads() -> Tuple[DownloadSet, DownloadSet]:
|
||||
model_sources = {}
|
||||
model_set = create_static_model_set('full')
|
||||
|
||||
for face_detector_model in [ 'retinaface', 'scrfd', 'yoloface' ]:
|
||||
for face_detector_model in [ 'retinaface', 'scrfd', 'yolo_face' ]:
|
||||
if state_manager.get_item('face_detector_model') in [ 'many', face_detector_model ]:
|
||||
model_hashes[face_detector_model] = model_set.get(face_detector_model).get('hashes').get(face_detector_model)
|
||||
model_sources[face_detector_model] = model_set.get(face_detector_model).get('sources').get(face_detector_model)
|
||||
@@ -122,8 +122,8 @@ def detect_faces(vision_frame : VisionFrame) -> Tuple[List[BoundingBox], List[Sc
|
||||
all_face_scores.extend(face_scores)
|
||||
all_face_landmarks_5.extend(face_landmarks_5)
|
||||
|
||||
if state_manager.get_item('face_detector_model') in [ 'many', 'yoloface' ]:
|
||||
bounding_boxes, face_scores, face_landmarks_5 = detect_with_yoloface(vision_frame, state_manager.get_item('face_detector_size'))
|
||||
if state_manager.get_item('face_detector_model') in [ 'many', 'yolo_face' ]:
|
||||
bounding_boxes, face_scores, face_landmarks_5 = detect_with_yolo_face(vision_frame, state_manager.get_item('face_detector_size'))
|
||||
all_bounding_boxes.extend(bounding_boxes)
|
||||
all_face_scores.extend(face_scores)
|
||||
all_face_landmarks_5.extend(face_landmarks_5)
|
||||
@@ -149,6 +149,7 @@ def detect_with_retinaface(vision_frame : VisionFrame, face_detector_size : str)
|
||||
feature_strides = [ 8, 16, 32 ]
|
||||
feature_map_channel = 3
|
||||
anchor_total = 2
|
||||
face_detector_score = state_manager.get_item('face_detector_score')
|
||||
face_detector_width, face_detector_height = unpack_resolution(face_detector_size)
|
||||
temp_vision_frame = resize_frame_resolution(vision_frame, (face_detector_width, face_detector_height))
|
||||
ratio_height = vision_frame.shape[0] / temp_vision_frame.shape[0]
|
||||
@@ -157,29 +158,29 @@ def detect_with_retinaface(vision_frame : VisionFrame, face_detector_size : str)
|
||||
detection = forward_with_retinaface(detect_vision_frame)
|
||||
|
||||
for index, feature_stride in enumerate(feature_strides):
|
||||
keep_indices = numpy.where(detection[index] >= state_manager.get_item('face_detector_score'))[0]
|
||||
keep_indices = numpy.where(detection[index] >= face_detector_score)[0]
|
||||
|
||||
if numpy.any(keep_indices):
|
||||
stride_height = face_detector_height // feature_stride
|
||||
stride_width = face_detector_width // feature_stride
|
||||
anchors = create_static_anchors(feature_stride, anchor_total, stride_height, stride_width)
|
||||
bounding_box_raw = detection[index + feature_map_channel] * feature_stride
|
||||
face_landmark_5_raw = detection[index + feature_map_channel * 2] * feature_stride
|
||||
bounding_boxes_raw = detection[index + feature_map_channel] * feature_stride
|
||||
face_landmarks_5_raw = detection[index + feature_map_channel * 2] * feature_stride
|
||||
|
||||
for bounding_box in distance_to_bounding_box(anchors, bounding_box_raw)[keep_indices]:
|
||||
for bounding_box_raw in distance_to_bounding_box(anchors, bounding_boxes_raw)[keep_indices]:
|
||||
bounding_boxes.append(numpy.array(
|
||||
[
|
||||
bounding_box[0] * ratio_width,
|
||||
bounding_box[1] * ratio_height,
|
||||
bounding_box[2] * ratio_width,
|
||||
bounding_box[3] * ratio_height,
|
||||
bounding_box_raw[0] * ratio_width,
|
||||
bounding_box_raw[1] * ratio_height,
|
||||
bounding_box_raw[2] * ratio_width,
|
||||
bounding_box_raw[3] * ratio_height
|
||||
]))
|
||||
|
||||
for score in detection[index][keep_indices]:
|
||||
face_scores.append(score[0])
|
||||
for face_score_raw in detection[index][keep_indices]:
|
||||
face_scores.append(face_score_raw[0])
|
||||
|
||||
for face_landmark_5 in distance_to_face_landmark_5(anchors, face_landmark_5_raw)[keep_indices]:
|
||||
face_landmarks_5.append(face_landmark_5 * [ ratio_width, ratio_height ])
|
||||
for face_landmark_raw_5 in distance_to_face_landmark_5(anchors, face_landmarks_5_raw)[keep_indices]:
|
||||
face_landmarks_5.append(face_landmark_raw_5 * [ ratio_width, ratio_height ])
|
||||
|
||||
return bounding_boxes, face_scores, face_landmarks_5
|
||||
|
||||
@@ -191,6 +192,7 @@ def detect_with_scrfd(vision_frame : VisionFrame, face_detector_size : str) -> T
|
||||
feature_strides = [ 8, 16, 32 ]
|
||||
feature_map_channel = 3
|
||||
anchor_total = 2
|
||||
face_detector_score = state_manager.get_item('face_detector_score')
|
||||
face_detector_width, face_detector_height = unpack_resolution(face_detector_size)
|
||||
temp_vision_frame = resize_frame_resolution(vision_frame, (face_detector_width, face_detector_height))
|
||||
ratio_height = vision_frame.shape[0] / temp_vision_frame.shape[0]
|
||||
@@ -199,65 +201,66 @@ def detect_with_scrfd(vision_frame : VisionFrame, face_detector_size : str) -> T
|
||||
detection = forward_with_scrfd(detect_vision_frame)
|
||||
|
||||
for index, feature_stride in enumerate(feature_strides):
|
||||
keep_indices = numpy.where(detection[index] >= state_manager.get_item('face_detector_score'))[0]
|
||||
keep_indices = numpy.where(detection[index] >= face_detector_score)[0]
|
||||
|
||||
if numpy.any(keep_indices):
|
||||
stride_height = face_detector_height // feature_stride
|
||||
stride_width = face_detector_width // feature_stride
|
||||
anchors = create_static_anchors(feature_stride, anchor_total, stride_height, stride_width)
|
||||
bounding_box_raw = detection[index + feature_map_channel] * feature_stride
|
||||
face_landmark_5_raw = detection[index + feature_map_channel * 2] * feature_stride
|
||||
bounding_boxes_raw = detection[index + feature_map_channel] * feature_stride
|
||||
face_landmarks_5_raw = detection[index + feature_map_channel * 2] * feature_stride
|
||||
|
||||
for bounding_box in distance_to_bounding_box(anchors, bounding_box_raw)[keep_indices]:
|
||||
for bounding_box_raw in distance_to_bounding_box(anchors, bounding_boxes_raw)[keep_indices]:
|
||||
bounding_boxes.append(numpy.array(
|
||||
[
|
||||
bounding_box[0] * ratio_width,
|
||||
bounding_box[1] * ratio_height,
|
||||
bounding_box[2] * ratio_width,
|
||||
bounding_box[3] * ratio_height,
|
||||
bounding_box_raw[0] * ratio_width,
|
||||
bounding_box_raw[1] * ratio_height,
|
||||
bounding_box_raw[2] * ratio_width,
|
||||
bounding_box_raw[3] * ratio_height
|
||||
]))
|
||||
|
||||
for score in detection[index][keep_indices]:
|
||||
face_scores.append(score[0])
|
||||
for face_score_raw in detection[index][keep_indices]:
|
||||
face_scores.append(face_score_raw[0])
|
||||
|
||||
for face_landmark_5 in distance_to_face_landmark_5(anchors, face_landmark_5_raw)[keep_indices]:
|
||||
face_landmarks_5.append(face_landmark_5 * [ ratio_width, ratio_height ])
|
||||
for face_landmark_raw_5 in distance_to_face_landmark_5(anchors, face_landmarks_5_raw)[keep_indices]:
|
||||
face_landmarks_5.append(face_landmark_raw_5 * [ ratio_width, ratio_height ])
|
||||
|
||||
return bounding_boxes, face_scores, face_landmarks_5
|
||||
|
||||
|
||||
def detect_with_yoloface(vision_frame : VisionFrame, face_detector_size : str) -> Tuple[List[BoundingBox], List[Score], List[FaceLandmark5]]:
|
||||
def detect_with_yolo_face(vision_frame : VisionFrame, face_detector_size : str) -> Tuple[List[BoundingBox], List[Score], List[FaceLandmark5]]:
|
||||
bounding_boxes = []
|
||||
face_scores = []
|
||||
face_landmarks_5 = []
|
||||
face_detector_score = state_manager.get_item('face_detector_score')
|
||||
face_detector_width, face_detector_height = unpack_resolution(face_detector_size)
|
||||
temp_vision_frame = resize_frame_resolution(vision_frame, (face_detector_width, face_detector_height))
|
||||
ratio_height = vision_frame.shape[0] / temp_vision_frame.shape[0]
|
||||
ratio_width = vision_frame.shape[1] / temp_vision_frame.shape[1]
|
||||
detect_vision_frame = prepare_detect_frame(temp_vision_frame, face_detector_size)
|
||||
detection = forward_with_yoloface(detect_vision_frame)
|
||||
detection = forward_with_yolo_face(detect_vision_frame)
|
||||
detection = numpy.squeeze(detection).T
|
||||
bounding_box_raw, score_raw, face_landmark_5_raw = numpy.split(detection, [ 4, 5 ], axis = 1)
|
||||
keep_indices = numpy.where(score_raw > state_manager.get_item('face_detector_score'))[0]
|
||||
bounding_boxes_raw, face_scores_raw, face_landmarks_5_raw = numpy.split(detection, [ 4, 5 ], axis = 1)
|
||||
keep_indices = numpy.where(face_scores_raw > face_detector_score)[0]
|
||||
|
||||
if numpy.any(keep_indices):
|
||||
bounding_box_raw, face_landmark_5_raw, score_raw = bounding_box_raw[keep_indices], face_landmark_5_raw[keep_indices], score_raw[keep_indices]
|
||||
bounding_boxes_raw, face_scores_raw, face_landmarks_5_raw = bounding_boxes_raw[keep_indices], face_scores_raw[keep_indices], face_landmarks_5_raw[keep_indices]
|
||||
|
||||
for bounding_box in bounding_box_raw:
|
||||
for bounding_box_raw in bounding_boxes_raw:
|
||||
bounding_boxes.append(numpy.array(
|
||||
[
|
||||
(bounding_box[0] - bounding_box[2] / 2) * ratio_width,
|
||||
(bounding_box[1] - bounding_box[3] / 2) * ratio_height,
|
||||
(bounding_box[0] + bounding_box[2] / 2) * ratio_width,
|
||||
(bounding_box[1] + bounding_box[3] / 2) * ratio_height,
|
||||
(bounding_box_raw[0] - bounding_box_raw[2] / 2) * ratio_width,
|
||||
(bounding_box_raw[1] - bounding_box_raw[3] / 2) * ratio_height,
|
||||
(bounding_box_raw[0] + bounding_box_raw[2] / 2) * ratio_width,
|
||||
(bounding_box_raw[1] + bounding_box_raw[3] / 2) * ratio_height
|
||||
]))
|
||||
|
||||
face_scores = score_raw.ravel().tolist()
|
||||
face_landmark_5_raw[:, 0::3] = (face_landmark_5_raw[:, 0::3]) * ratio_width
|
||||
face_landmark_5_raw[:, 1::3] = (face_landmark_5_raw[:, 1::3]) * ratio_height
|
||||
face_scores = face_scores_raw.ravel().tolist()
|
||||
face_landmarks_5_raw[:, 0::3] = (face_landmarks_5_raw[:, 0::3]) * ratio_width
|
||||
face_landmarks_5_raw[:, 1::3] = (face_landmarks_5_raw[:, 1::3]) * ratio_height
|
||||
|
||||
for face_landmark_5 in face_landmark_5_raw:
|
||||
face_landmarks_5.append(numpy.array(face_landmark_5.reshape(-1, 3)[:, :2]))
|
||||
for face_landmark_raw_5 in face_landmarks_5_raw:
|
||||
face_landmarks_5.append(numpy.array(face_landmark_raw_5.reshape(-1, 3)[:, :2]))
|
||||
|
||||
return bounding_boxes, face_scores, face_landmarks_5
|
||||
|
||||
@@ -286,8 +289,8 @@ def forward_with_scrfd(detect_vision_frame : VisionFrame) -> Detection:
|
||||
return detection
|
||||
|
||||
|
||||
def forward_with_yoloface(detect_vision_frame : VisionFrame) -> Detection:
|
||||
face_detector = get_inference_pool().get('yoloface')
|
||||
def forward_with_yolo_face(detect_vision_frame : VisionFrame) -> Detection:
|
||||
face_detector = get_inference_pool().get('yolo_face')
|
||||
|
||||
with thread_semaphore():
|
||||
detection = face_detector.run(None,
|
||||
|
||||
@@ -94,7 +94,7 @@ def create_output_pattern_program() -> ArgumentParser:
|
||||
def create_face_detector_program() -> ArgumentParser:
|
||||
program = ArgumentParser(add_help = False)
|
||||
group_face_detector = program.add_argument_group('face detector')
|
||||
group_face_detector.add_argument('--face-detector-model', help = wording.get('help.face_detector_model'), default = config.get_str_value('face_detector.face_detector_model', 'yoloface'), choices = facefusion.choices.face_detector_models)
|
||||
group_face_detector.add_argument('--face-detector-model', help = wording.get('help.face_detector_model'), default = config.get_str_value('face_detector.face_detector_model', 'yolo_face'), choices = facefusion.choices.face_detector_models)
|
||||
known_args, _ = program.parse_known_args()
|
||||
face_detector_size_choices = facefusion.choices.face_detector_set.get(known_args.face_detector_model)
|
||||
group_face_detector.add_argument('--face-detector-size', help = wording.get('help.face_detector_size'), default = config.get_str_value('face_detector.face_detector_size', get_last(face_detector_size_choices)), choices = face_detector_size_choices)
|
||||
|
||||
@@ -98,7 +98,7 @@ LogLevelSet = Dict[LogLevel, int]
|
||||
TableHeaders = List[str]
|
||||
TableContents = List[List[Any]]
|
||||
|
||||
FaceDetectorModel = Literal['many', 'retinaface', 'scrfd', 'yoloface']
|
||||
FaceDetectorModel = Literal['many', 'retinaface', 'scrfd', 'yolo_face']
|
||||
FaceLandmarkerModel = Literal['many', '2dfan4', 'peppa_wutz']
|
||||
FaceDetectorSet = Dict[FaceDetectorModel, List[str]]
|
||||
FaceSelectorMode = Literal['many', 'one', 'reference']
|
||||
|
||||
Reference in New Issue
Block a user