def face_detect(file, frame_path, csv_data):
name = file.replace('.jpg', '').split('-')
log = csv_data.iloc[int(name[0])]
x = log[3]
y = log[4]
img = cv2.imread('%s/%s' % (frame_path, file))
img_ = Image.open('%s/%s' % (frame_path, file))
x = img.shape[1] * x
y = img.shape[0] * y
faces, landmarks = detector.detect_faces(img_)
# check if detected faces
if len(faces) == 0:
# print('no face detect: ' + file)
return # no face
faces[0][2] = faces[0][2] - faces[0][0]
faces[0][3] = faces[0][3] - faces[0][1]
bounding_box = bounding_box_check(faces, x, y)
if bounding_box is None:
# print('face is not related to given coord: ' + file)
return
# print(file, " ", bounding_box)
# print(file, " ", x, y)
crop_img = img[bounding_box[1]:bounding_box[1] + bounding_box[3],
bounding_box[0]:bounding_box[0] + bounding_box[2]]
crop_img = cv2.resize(crop_img, (160, 160))
cv2.imwrite('%s/frame_' % output_dir + name[0] + '_' + name[1] + '.jpg',
crop_img)
def predict():
files = request.files
img_left = Image.open(files.get('imgLeft')).convert('RGB')
img_right = Image.open(files.get('imgRight')).convert('RGB')
bbox_left, _ = detect_faces(img_left)
bbox_right, _ = detect_faces(img_right)
if bbox_left.shape[0] > 0:
a, b, c, d, _ = bbox_left[0]
img_left = img_left.crop((a, b, c, d))
if bbox_right.shape[0] > 0:
a, b, c, d, _ = bbox_right[0]
img_right = img_right.crop((a, b, c, d))
distance, similar = is_same(img_left, img_right, THRESHOLD)
model_acc = ModelLoaded.acc
return jsonify(same=('BERBEDA', 'SAMA')[similar.item()],
score=distance.item(),
model_acc=model_acc,
threshold=THRESHOLD)
def fetch_faces(image, return_landmarks=False):
# standarize detector input from cv2 image to PIL Image
if isinstance(image, (np.ndarray, np.generic)):
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
image = Image.fromarray(image)
# for some reason, detector randomly throws an error
try:
bboxes, landmarks = detect_faces(image)
except ValueError:
bboxes, landmarks = [], []
# postprocess bounding bboxes
if len(bboxes) > 0:
scores = bboxes[:, -1]
bboxes = bboxes[:, :-1].astype("int")
else:
scores = []
return ([bboxes, scores],
landmarks) if return_landmarks else bboxes, scores
def detect_and_store(path, final_root_dir, new_size):
img = Image.open(path)
bounding_boxes, landmarks = detect_faces(img)
if len(bounding_boxes) == 0:
print("No face detected on image", get_dir_and_file(path))
'''
image should only contains one valid person for the corresponding label/person
so, we assume that bounding_boxes shape is always (1, 5)
'''
for a, b, c, d, _ in bounding_boxes:
dst = os.path.join(final_root_dir, path.parent.name)
os.makedirs(dst, exist_ok=True)
final_name = os.path.join(dst, path.name)
try:
img.crop((a, b, c, d)).resize((new_size, new_size),
Image.BILINEAR).save(final_name)
print(get_dir_and_file(path), 'saved to', final_name)
except Exception as e:
print("Error occured when saving", get_dir_and_file(path))
print("Error: ", str(e))
from torch_mtcnn import detect_faces
from PIL import Image
from os import listdir
from os.path import isfile, join
import json
import csv
with open('configuration.json', 'r') as configfile:
config_info=json.load(configfile)
print(config_info['input-images-directory'])
input_images_path=config_info['input-images-directory']
print(input_images_path)
image_files = [f for f in listdir(input_images_path) if isfile(join(input_images_path, f))]
cropped_images_path=config_info['cropped-faces-directory']
for file in image_files:
image = Image.open(input_images_path+file).convert('RGB')
bounding_boxes, landmarks = detect_faces(image)
for i in range(len(bounding_boxes)):
area = (bounding_boxes[i][0],bounding_boxes[i][1],bounding_boxes[i][2],bounding_boxes[i][3])
t_image=image.crop(area)
j=i+1
t_image=t_image.save(cropped_images_path+str(j)+'-'+file)
print(m.keys())
classes = m['classes']
net = eval(args.model.type)(args.model.num_classes)
net.load_state_dict(m['model'])
net.to(torch.device('cuda'))
net.eval()
i = 0
while ret:
raw = frame
i+=1
if i % 2 == 0:
continue
pil_img = cv2_to_pil(raw)
bounding_boxes, landmarks = detect_faces(pil_img)
print(bounding_boxes)
for bbox in bounding_boxes:
x1, y1, x2, y2 = int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])
cv2.rectangle(raw, (x1, y1), (x2, y2), (0, 255, 0))
frame = raw[y1: y2, x1: x2, :]
cv2.imshow('face', frame)
# 模型处理
img = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
img = cv2.resize(img, (224, 224))