def detect_faces_batch(images):
detector = face_detection.FaceAlignment(face_detection.LandmarksType._2D,
flip_input=False,
device=device)
batch_size = 32
while 1:
predictions = []
try:
for i in range(0, len(images), batch_size):
predictions.extend(
detector.get_detections_for_batch(
np.array(images[i:i + batch_size]), True))
except RuntimeError:
if batch_size == 1:
raise RuntimeError(
'Image too big to run face detection on GPU. Please use the --resize_factor argument'
)
batch_size //= 2
print('Recovering from OOM error; New batch size: {}'.format(
batch_size))
continue
break
del detector
return predictions
def face_detect(images):
if args.cartoon:
model = insightface.model_zoo.get_model('retinaface_r50_v1')
model.prepare(ctx_id=-1, nms=0.4)
predictions = []
for image in images:
bbox, landmark = model.detect(image, threshold=0.3, scale=1.0)
bbox = np.array(bbox).astype(np.int)[:, :-1]
predictions.append(tuple(bbox[0].tolist()))
else:
detector = face_detection.FaceAlignment(
face_detection.LandmarksType._2D, flip_input=False, device=device)
batch_size = args.face_det_batch_size
while 1:
predictions = []
try:
for i in tqdm(range(0, len(images), batch_size)):
predictions.extend(
detector.get_detections_for_batch(
np.array(images[i:i + batch_size])))
except RuntimeError:
if batch_size == 1:
raise RuntimeError(
'Image too big to run face detection on GPU. Please use the --resize_factor argument'
)
batch_size //= 2
print('Recovering from OOM error; New batch size: {}'.format(
batch_size))
continue
break
results = []
pady1, pady2, padx1, padx2 = args.pads
for rect, image in zip(predictions, images):
if rect is None:
cv2.imwrite(
'temp/faulty_frame.jpg',
image) # check this frame where the face was not detected.
raise ValueError(
'Face not detected! Ensure the video contains a face in all the frames.'
)
y1 = max(0, rect[1] - pady1)
y2 = min(image.shape[0], rect[3] + pady2)
x1 = max(0, rect[0] - padx1)
x2 = min(image.shape[1], rect[2] + padx2)
results.append([x1, y1, x2, y2])
boxes = np.array(results)
if not args.nosmooth: boxes = get_smoothened_boxes(boxes, T=5)
results = [[image[y1:y2, x1:x2], (y1, y2, x1, x2)]
for image, (x1, y1, x2, y2) in zip(images, boxes)]
if not args.cartoon:
del detector
return results
def face_detect(images):
detector = face_detection.FaceAlignment(face_detection.LandmarksType._2D,
flip_input=False,
device=device)
batch_size = args.face_det_batch_size
while 1:
predictions = []
try:
for i in tqdm(range(0, len(images), batch_size)):
predictions.extend(
detector.get_detections_for_batch(
np.array(images[i:i + batch_size])))
except RuntimeError:
if batch_size == 1:
raise RuntimeError(
'Image too big to run face detection on GPU. Please use the --resize_factor argument'
)
batch_size //= 2
print('Recovering from OOM error; New batch size: {}'.format(
batch_size))
continue
break
results = []
pady1, pady2, padx1, padx2 = args.pads
for rect, image in zip(predictions, images):
if rect is None and args.ignorefaceless:
y1 = max(0, 0)
y2 = min(0, 0)
x1 = max(0, 0)
x2 = min(0, 0)
elif rect is None:
cv2.imwrite(
'temp/faulty_frame.jpg',
image) # check this frame where the face was not detected.
raise ValueError(
'Face not detected! Ensure the video contains a face in all the frames.'
)
else:
y1 = max(0, rect[1] - pady1)
y2 = min(image.shape[0], rect[3] + pady2)
x1 = max(0, rect[0] - padx1)
x2 = min(image.shape[1], rect[2] + padx2)
results.append([x1, y1, x2, y2])
boxes = np.array(results)
if not args.nosmooth: boxes = get_smoothened_boxes(boxes, T=5)
results = [[image[y1:y2, x1:x2], (y1, y2, x1, x2)]
for image, (x1, y1, x2, y2) in zip(images, boxes)]
del detector
return results
def face_detect_rect_first_one_only(images_ori):
nosmooth = False
pads = [0, 0, 0, 0]
face_det_batch_size = 1
images = images_ori[:1]
detector = face_detection.FaceAlignment(face_detection.LandmarksType._2D,
flip_input=False,
device=device)
batch_size = face_det_batch_size
while 1:
predictions = []
try:
for i in range(0, len(images), batch_size):
predictions.extend(
detector.get_detections_for_batch(
np.array(images[i:i + batch_size])))
except RuntimeError:
if batch_size == 1:
raise RuntimeError(
'Image too big to run face detection on GPU. Please use the --resize_factor argument'
)
batch_size //= 2
print('Recovering from OOM error; New batch size: {}'.format(
batch_size))
continue
break
results = []
pady1, pady2, padx1, padx2 = pads
for rect, image in zip(predictions, images):
if rect is None:
# check this frame where the face was not detected.
# cv2.imwrite('temp/faulty_frame.jpg', image)
print(
'Face not detected! Ensure the video contains a face in all the frames.'
)
return []
y1 = max(0, rect[1] - pady1)
y2 = min(image.shape[0], rect[3] + pady2)
x1 = max(0, rect[0] - padx1)
x2 = min(image.shape[1], rect[2] + padx2)
# y_gap, x_gap = (y2 - y1)//2, (x2 - x1)//2
y_gap = min((y2 - y1) // 2, y1, image.shape[0] - y2)
x_gap = (((y2 - y1) + y_gap * 2) - (x2 - x1)) // 2
coords = y1 - y_gap, y2 + y_gap, x1 - x_gap, x2 + x_gap
#print(coords)
# coords = [coords_[0], coords_[0]+1024, coords_[2], coords_[2]+1024]
#results.append(image[y1-y_gap: y2+y_gap, x1-x_gap:x2+x_gap])
results = [
image[coords[0]:coords[1], coords[2]:coords[3]] for image in images_ori
]
del detector
return results
def detect_faces(image):
detector = face_detection.FaceAlignment(face_detection.LandmarksType._2D,
flip_input=False,
device=device)
predictions = []
while 1:
try:
predictions.extend(
detector.get_detections_for_batch(np.array(image), False))
break
except:
pass
del detector
return predictions
from glob import glob
import audio
from hparams import hparams as hp
import face_detection
parser = argparse.ArgumentParser()
parser.add_argument('--ngpu', help='Number of GPUs across which to run in parallel', default=1, type=int)
parser.add_argument('--batch_size', help='Single GPU Face detection batch size', default=32, type=int)
parser.add_argument("--data_root", help="Root folder of the LRS2 dataset", required=True)
parser.add_argument("--preprocessed_root", help="Root folder of the preprocessed dataset", required=True)
args = parser.parse_args()
fa = [face_detection.FaceAlignment(face_detection.LandmarksType._2D, flip_input=False,
device='cuda:{}'.format(id)) for id in range(args.ngpu)]
template = 'ffmpeg -loglevel panic -y -i {} -strict -2 {}'
# template2 = 'ffmpeg -hide_banner -loglevel panic -threads 1 -y -i {} -async 1 -ac 1 -vn -acodec pcm_s16le -ar 16000 {}'
def process_video_file(vfile, args, gpu_id):
video_stream = cv2.VideoCapture(vfile)
frames = []
while 1:
still_reading, frame = video_stream.read()
if not still_reading:
video_stream.release()
break
frames.append(frame)
import torch, face_detection
import scipy, cv2, os, sys, argparse, audio
import numpy as np
from tqdm import tqdm
face = '../videos/main/fps_corrected_video/video-3-0-4a.mp4'
video_stream = cv2.VideoCapture(face)
fps = video_stream.get(cv2.CAP_PROP_FPS)
crop = [0, -1, 0, -1]
print('Reading video frames...')
device = 'cuda' if torch.cuda.is_available() else 'cpu'
detector = face_detection.FaceAlignment(face_detection.LandmarksType._2D,
flip_input=False,
device='cpu')
full_frames = []
while 1:
still_reading, frame = video_stream.read()
if not still_reading:
video_stream.release()
break
y1, y2, x1, x2 = crop
if x2 == -1: x2 = frame.shape[1]
if y2 == -1: y2 = frame.shape[0]
frame = frame[y1:y2, x1:x2]
full_frames.append(frame)
predictions = []
parser.add_argument('--batch_size',
help='Single GPU Face detection batch size',
default=32,
type=int)
parser.add_argument("--data_root",
help="Root folder of the LRS2 dataset",
required=True)
parser.add_argument("--preprocessed_root",
help="Root folder of the preprocessed dataset",
required=True)
args = parser.parse_args()
fa = [
face_detection.FaceAlignment(face_detection.LandmarksType._2D,
flip_input=False,
device='cuda:{}'.format(id))
for id in range(args.ngpu)
]
template = 'ffmpeg -loglevel panic -y -i {} -strict -2 {}'
# template2 = 'ffmpeg -hide_banner -loglevel panic -threads 1 -y -i {} -async 1 -ac 1 -vn -acodec pcm_s16le -ar 16000 {}'
def process_video_file(vfile, args, gpu_id):
video_stream = cv2.VideoCapture(vfile)
frames = []
while 1:
still_reading, frame = video_stream.read()
if not still_reading:
def face_detect_rect(images, images_ori):
nosmooth = False
pads = [0, 0, 0, 0]
face_det_batch_size = 32
detector = face_detection.FaceAlignment(face_detection.LandmarksType._2D,
flip_input=False,
device=device)
batch_size = face_det_batch_size
while 1:
predictions = []
try:
for i in range(0, len(images), batch_size):
predictions.extend(
detector.get_detections_for_batch(
np.array(images[i:i + batch_size])))
except RuntimeError:
if batch_size == 1:
raise RuntimeError(
'Image too big to run face detection on GPU. Please use the --resize_factor argument'
)
batch_size //= 2
print('Recovering from OOM error; New batch size: {}'.format(
batch_size))
continue
break
all_coords = []
results = []
pady1, pady2, padx1, padx2 = pads
for rect, image in zip(predictions, images):
if rect is None:
# check this frame where the face was not detected.
cv2.imwrite('temp/faulty_frame.jpg', image)
print(
'Face not detected! Ensure the video contains a face in all the frames.'
)
return []
y1 = max(0, rect[1] - pady1)
y2 = min(image.shape[0], rect[3] + pady2)
x1 = max(0, rect[0] - padx1)
x2 = min(image.shape[1], rect[2] + padx2)
y_gap, x_gap = (y2 - y1) // 6, (x2 - x1) // 6
all_coords.append(np.array([y1, y2, x1, x2]))
results.append(image[y1 - y_gap:y2 + y_gap, x1 - x_gap:x2 + x_gap])
var = np.var(all_coords, axis=0)
del detector
if sum(var) > 100:
results = [cv2.resize(img, (128, 128)) for img in results]
else:
first_rect = predictions[0]
y1 = max(0, rect[1] - pady1)
y2 = min(image.shape[0], rect[3] + pady2)
x1 = max(0, rect[0] - padx1)
x2 = min(image.shape[1], rect[2] + padx2)
y_gap, x_gap = (y2 - y1) // 6, (x2 - x1) // 6
results = [
cv2.resize(img[y1 - y_gap:y2 + y_gap, x1 - x_gap:x2 + x_gap],
(128, 128)) for img in images
]
return results
def face_detect(images, video):
if not args.cache:
detector = face_detection.FaceAlignment(
face_detection.LandmarksType._2D, flip_input=False, device=device)
batch_size = args.face_det_batch_size
while 1:
predictions = []
try:
for i in tqdm(range(0, len(images), batch_size)):
predictions.extend(
detector.get_detections_for_batch(
np.array(images[i:i + batch_size]), video))
except RuntimeError:
if batch_size == 1:
raise RuntimeError(
'Image too big to run face detection on GPU. Please use the --resize_factor argument'
)
batch_size //= 2
print('Recovering from OOM error; New batch size: {}'.format(
batch_size))
continue
del detector
break
else:
df = pd.read_csv(args.cache)
preds = df.values.tolist()
if len(preds) == 1:
predictions = [[eval(pred) for pred in preds[0]]]
elif len(preds) > 1:
predictions = [[eval(pred)] for batch in preds for pred in batch]
if args.static or args.static_video:
face = predictions[0][args.person]
predictions = [face]
else:
tmp = []
for pred in predictions:
face = pred[0]
tmp.append(face)
predictions = tmp
results = []
pady1, pady2, padx1, padx2 = args.pads
for rect, image in zip(predictions, images):
if rect is None:
cv2.imwrite(
'temp/faulty_frame.jpg',
image) # check this frame where the face was not detected.
raise ValueError(
'Face not detected! Ensure the video contains a face in all the frames.'
)
y1 = max(0, rect[1] - pady1)
y2 = min(image.shape[0], rect[3] + pady2)
x1 = max(0, rect[0] - padx1)
x2 = min(image.shape[1], rect[2] + padx2)
results.append([x1, y1, x2, y2])
boxes = np.array(results)
if not args.nosmooth: boxes = get_smoothened_boxes(boxes, T=5)
results = [[image[y1:y2, x1:x2], (y1, y2, x1, x2)]
for image, (x1, y1, x2, y2) in zip(images, boxes)]
return results
import dlib
import sys
import os
import face_detection
# print("Dlib using cuda?")
# print(dlib.DLIB_USE_CUDA)
face_detector_path = sys.argv[1]
frames_path = sys.argv[2]
save_path = sys.argv[3]
#cnn_face_detector = dlib.cnn_face_detection_model_v1(face_detector_path)
fa = face_detection.FaceAlignment(face_detection.LandmarksType._2D)
print('Reading video frames...')
numImg = len([
name for name in os.listdir(frames_path)
if os.path.isfile(os.path.join(frames_path, name))
])
from_first = True
with open(save_path, 'w+') as out:
for f in range(1, numImg + 1):
number = '{0:04d}'.format(f)
filename = os.path.join(frames_path, "frames" + number + ".jpg")
img = dlib.load_rgb_image(filename)
dets = fa.get_detections_for_image(img)