def getDataFromTXT_68(filepath, test=False):
'''
Get data from dataset mentioned in paper.
Input:
- filepath: trainImageList or testImageList
Output:
- A tuple of (imgpath, bbox, landmark)
- imgpath: train image or test image
- bbox: type of BBox
- landmark: (5L, 2L) of [0,1]
'''
dirname = os.path.dirname(filepath)
f = open(filepath, 'r')
data = []
for line in f.readlines():
s = line.strip().split(' ')
imgPath = os.path.join(dirname, s[0].replace('\\', '/'))
bbox = map(int, [s[1], s[2], s[3], s[4]])
bbox = BBox(bbox)
if test:
data.append((imgPath, bbox))
continue
landmark = np.zeros((68, 2))
for i in range(0, 68):
landmark[i] = (float(s[5 + i * 2]), float(s[5 + i * 2 + 1]))
landmark = bbox.projectLandmark(landmark) #[0,1]
data.append((imgPath, bbox, landmark))
return data
def getDataFromTXT(filepath, test=False):
'''
Get data from dataset mentioned in paper.
Input:
- filepath: trainImageList or testImageList
Output:
- A tuple of (imgpath, bbox, landmark)
- imgpath: train image or test image
- bbox: type of BBox
- landmark: (5L, 2L) of [0,1]
'''
dirname = os.path.dirname(filepath)
f = open(filepath, 'r')
data = []
for line in f.readlines():
s = line.strip().split(' ')
imgPath = os.path.join(dirname, s[0].replace('\\', '/'))
bbox = map(int, [s[1], s[2], s[3], s[4]])
bbox = BBox(bbox)
if test:
data.append((imgPath, bbox))
continue
landmark = np.zeros((5,2))
for i in range(0,5):
landmark[i] = (float(s[5+i*2]), float(s[5+i*2+1]))
landmark = bbox.projectLandmark(landmark) #[0,1]
data.append((imgPath, bbox, landmark))
return data
def load_celeba_data():
'''
load celeba dataset and crop the face box
Return a tuple of:
- img_path: dataset/celeba/000001.jpg
- bbox: object of BBox
- landmark: (5L, 2L) of [0,1]
'''
text = '/home/cunjian/code/caffe/examples/dataset/celeba/list_landmarks_celeba.txt'
# text = 'E:\\dataset\\CelebA\\list_landmarks_celeba.txt'
fin = open(text, 'r')
n = int(fin.readline().strip())
fin.readline() # drop this line [lefteye_x, lefteye_y, ...]
result = []
for i in range(n):
line = fin.readline().strip()
components = line.split()
img_path = join('../dataset/img_celeba', components[0])
# img_path = join('E:\\dataset\\CelebA\\img_celeba', components[0])
landmark = np.asarray([int(value) for value in components[1:]],
dtype=np.float32)
landmark = landmark.reshape(len(landmark) / 2, 2)
# crop face box
x_max, y_max = landmark.max(0)
x_min, y_min = landmark.min(0)
w, h = x_max - x_min, y_max - y_min
w = h = min(w, h)
ratio = 0 # default 0.5
x_new = x_min - w * ratio
y_new = y_min - h * ratio
w_new = w * (1 + 2 * ratio)
h_new = h * (1 + 2 * ratio)
bbox = map(int, [x_new, x_new + w_new, y_new, y_new + h_new])
bbox = BBox(bbox)
# normalize landmark
landmark = bbox.projectLandmark(landmark)
#print landmark, if uncommented, the program will report error since some samples are omitted
#fit=0
#for index in range(0,5):
# if landmark[index,0]<0 or landmark[index,0]>1 or landmark[index,1]<0 or landmark[index,1]>1:
# fit=1
# break
#if fit==0:
# result.append((img_path, bbox, landmark))
result.append((img_path, bbox, landmark))
fin.close()
return result
def getDataFromTXT(filepath, test=False):
'''
Get data from dataset mentioned in paper.
Input:
- filepath: trainImageList or testImageList
Output:
- A tuple of (imgpath, bbox, landmark)
- imgpath: train image or test image
- bbox: type of BBox
- landmark: (5L, 2L) of [0,1]
'''
dirname = os.path.dirname(filepath)
f = open(filepath, 'r')
data = []
for line in f.readlines():
s = line.strip().split(' ')
imgPath = os.path.join(dirname, s[0].replace('\\', '/'))
bbox = map(int, [s[1], s[2], s[3], s[4]])
bbox = BBox(bbox)
if test:
x_max = bbox.right #bbox[1]
x_min = bbox.left #bbox[0]
y_max = bbox.bottom #bbox[3]
y_min = bbox.top #bbox[2]
# enlarge bounding box
w, h = x_max - x_min, y_max - y_min
w = h = min(w, h)
ratio = 0
x_new = x_min - w * ratio
y_new = y_min - h * ratio
w_new = w * (1 + 2 * ratio)
h_new = h * (1 + 2 * ratio)
new_bbox = map(int, [x_new, x_new + w_new, y_new, y_new + h_new])
new_bbox = BBox(new_bbox)
data.append((imgPath, new_bbox))
continue
landmark = np.zeros((5, 2))
for i in range(0, 5):
landmark[i] = (float(s[5 + i * 2]), float(s[5 + i * 2 + 1]))
landmark = bbox.projectLandmark(landmark) #[0,1]
#print landmark
fit = 0
for i in range(0, 5):
if landmark[i, 0] < 0 or landmark[i, 0] > 1 or landmark[
i, 1] < 0 or landmark[i, 1] > 1:
fit = 1
break
if fit == 0:
data.append((imgPath, bbox, landmark))
return data
def load_celeba_data():
'''
load celeba dataset and crop the face box
Return a tuple of:
- img_path: dataset/celeba/000001.jpg
- bbox: object of BBox
- landmark: (5L, 2L) of [0,1]
'''
text = '../dataset/celeba/list_landmarks_celeba.txt'
# text = 'E:\\dataset\\CelebA\\list_landmarks_celeba.txt'
fin = open(text, 'r')
n = int(fin.readline().strip())
fin.readline() # drop this line [lefteye_x, lefteye_y, ...]
result = []
for i in range(n):
line = fin.readline().strip()
components = line.split()
img_path = join('../dataset/celeba', components[0])
# img_path = join('E:\\dataset\\CelebA\\img_celeba', components[0])
landmark = np.asarray([int(value) for value in components[1:]],
dtype=np.float32)
landmark = landmark.reshape(len(landmark) / 2, 2)
# crop face box
x_max, y_max = landmark.max(0)
x_min, y_min = landmark.min(0)
w, h = x_max - x_min, y_max - y_min
w = h = min(w, h)
ratio = 0.5
x_new = x_min - w * ratio
y_new = y_min - h * ratio
w_new = w * (1 + 2 * ratio)
h_new = h * (1 + 2 * ratio)
bbox = map(int, [x_new, x_new + w_new, y_new, y_new + h_new])
bbox = BBox(bbox)
# normalize landmark
landmark = bbox.projectLandmark(landmark)
result.append((img_path, bbox, landmark))
fin.close()
return result
def load_celeba_data():
'''
load celeba dataset and crop the face box
Return a tuple of:
- img_path: dataset/celeba/000001.jpg
- bbox: object of BBox
- landmark: (5L, 2L) of [0,1]
'''
text = '../dataset/celeba/list_landmarks_celeba.txt'
# text = 'E:\\dataset\\CelebA\\list_landmarks_celeba.txt'
fin = open(text, 'r')
n = int(fin.readline().strip())
fin.readline() # drop this line [lefteye_x, lefteye_y, ...]
result = []
for i in range(n):
line = fin.readline().strip()
components = line.split()
img_path = join('../dataset/celeba', components[0])
# img_path = join('E:\\dataset\\CelebA\\img_celeba', components[0])
landmark = np.asarray([int(value) for value in components[1:]], dtype=np.float32)
landmark = landmark.reshape(len(landmark) / 2, 2)
# crop face box
x_max, y_max = landmark.max(0)
x_min, y_min = landmark.min(0)
w, h = x_max-x_min, y_max-y_min
w = h = min(w, h)
ratio = 0.5
x_new = x_min - w*ratio
y_new = y_min - h*ratio
w_new = w*(1 + 2*ratio)
h_new = h*(1 + 2*ratio)
bbox = map(int, [x_new, x_new+w_new, y_new, y_new+h_new])
bbox = BBox(bbox)
# normalize landmark
landmark = bbox.projectLandmark(landmark)
result.append((img_path, bbox, landmark))
fin.close()
return result
def get_fods_from_file(label_file):
"""
Get a list of fods from one label file.
:param label_file: Path of label xml file
:return: List of FODs from that xml file
"""
fods = []
frame_id = get_frame_id(label_file)
fod_labels = parse_label_file(label_file)
for fod_label in fod_labels:
bbox = BBox(fod_label['xmin'], fod_label['xmax'], fod_label['ymin'],
fod_label['ymax'])
fod_type = fod_label['name']
fod = FOD(frame_id, bbox, fod_type)
fods.append(fod)
return fods
def get_new_bbox_vals(bbox, rotate_degree):
c1_x = bbox.y_min * math.sin(rotate_degree) + bbox.x_min * math.cos(rotate_degree)
c1_y = bbox.y_min * math.cos(rotate_degree) + bbox.x_min * math.sin(rotate_degree)
c2_x = bbox.y_max * math.sin(rotate_degree) + bbox.x_min * math.cos(rotate_degree)
c2_y = bbox.y_max * math.cos(rotate_degree) + bbox.x_min * math.sin(rotate_degree)
c3_x = bbox.y_min * math.sin(rotate_degree) + bbox.x_max * math.cos(rotate_degree)
c3_y = bbox.y_min * math.cos(rotate_degree) + bbox.x_max * math.sin(rotate_degree)
c4_x = bbox.y_max * math.sin(rotate_degree) + bbox.x_max * math.cos(rotate_degree)
c4_y = bbox.y_max * math.cos(rotate_degree) + bbox.x_max * math.sin(rotate_degree)
x_min = int(min(c1_x, c2_x, c3_x, c4_x))
x_max = int(max(c1_x, c2_x, c3_x, c4_x))
y_min = int(min(c1_y, c2_y, c3_y, c4_y))
y_max = int(max(c1_y, c2_y, c3_y, c4_y))
bbox = BBox(x_min, x_max, y_min, y_max)
return bbox
x1 = cx - size // 2
x2 = x1 + size
y1 = cy - size // 2
y2 = y1 + size
dx = max(0, -x1)
dy = max(0, -y1)
x1 = max(0, x1)
y1 = max(0, y1)
edx = max(0, x2 - width)
edy = max(0, y2 - height)
x2 = min(width, x2)
y2 = min(height, y2)
new_bbox = list(map(int, [x1, x2, y1, y2]))
# new_bbox2 = [x1, x2, y1, y2]
new_bbox = BBox(new_bbox)
cropped = img[new_bbox.top:new_bbox.bottom, new_bbox.left:new_bbox.right]
cropped = cv2.copyMakeBorder(cropped, int(dy), int(edy), int(dx), int(edx),
cv2.BORDER_CONSTANT, 0)
cropped_face = cv2.resize(cropped, (out_size, out_size))
cropped_face = np.asarray(cv2.cvtColor(cropped_face, cv2.COLOR_BGR2RGB))
cropped_face = np.transpose(cropped_face, [2, 0, 1])
cv2.rectangle(image, (x1, y1), (x2, y2), (255, 0, 0), 2)
roi_color = image[x1:x1 + x2, y1:y1 + y2]
cropped_face = (np.asarray(cropped_face, dtype='float32')) / 255.0
# print(cropped_face.shape)
cropped_face = cropped_face.reshape(1, 3, 112, 112)
cropped_face = torch.from_numpy(cropped_face)
# print(cropped_face.shape)
def gen_lfw_landmark(img_dir, save_dir, input_size, argument, name):
# 数据输出路径
base_dir = os.path.join(save_dir, str(input_size))
if not os.path.exists(base_dir):
os.mkdir(base_dir)
landmark_save_dir = os.path.join(base_dir, '%s_landmark' % name)
if not os.path.exists(landmark_save_dir):
os.mkdir(landmark_save_dir)
# label记录txt
if name == 'lfw':
ftxt = os.path.join(img_dir, 'trainImageList.txt')
elif name == 'wider':
ftxt = os.path.join(img_dir, 'wider_face_train_landmark.txt')
img_dir = os.path.join(img_dir, 'WIDER_train/images')
else:
print('name只能是"lfw"或"wider"')
exit()
# 记录label的txt
f = open(os.path.join(base_dir, '%s_landmark.txt' % name), 'w')
# 获取图像路径, box, 关键点
data = get_data_from_txt(
ftxt, img_dir, name
) # lfw data format: [(path, BBox object, [[,], [,], [,], [,], [,]]), ]
idx = 0
image_id = 0
for (imgPath, box, landmarkGt) in tqdm(data):
# 存储人脸图片和关键点
F_imgs = []
F_landmarks = []
img = cv.imread(imgPath)
img_h, img_w, img_c = img.shape
gt_box = np.array([box.left, box.top, box.right, box.bottom])
# 人脸图片
f_face = img[box.top:box.bottom + 1, box.left:box.right + 1]
# resize成网络输入大小
f_face = cv.resize(f_face, (input_size, input_size))
landmark = np.zeros((5, 2))
for index, one in enumerate(landmarkGt):
# 关键点相对于左上坐标偏移量并归一化
rv = ((one[0] - gt_box[0]) / (gt_box[2] - gt_box[0]),
(one[1] - gt_box[1]) / (gt_box[3] - gt_box[1]))
landmark[index] = rv
F_imgs.append(f_face)
F_landmarks.append(landmark.reshape(10))
landmark = np.zeros((5, 2))
if argument:
# 对图像变换
idx = idx + 1
x1, y1, x2, y2 = gt_box
gt_w = x2 - x1 + 1
gt_h = y2 - y1 + 1
# 除去过小图像
if max(gt_w, gt_h) < 40 or x1 < 0 or y1 < 0:
continue
for i in range(10):
# 随机裁剪图像大小
box_size = np.random.randint(int(min(gt_w, gt_h) * 0.8),
np.ceil(1.25 * max(gt_w, gt_h)))
# 随机左上坐标偏移量
delta_x = np.random.randint(-gt_w * 0.2, gt_w * 0.2)
delta_y = np.random.randint(-gt_h * 0.2, gt_h * 0.2)
# 计算左上坐标
nx1 = int(max(x1 + gt_w / 2 - box_size / 2 + delta_x, 0))
ny1 = int(max(y1 + gt_h / 2 - box_size / 2 + delta_y, 0))
nx2 = nx1 + box_size
ny2 = ny1 + box_size
# 除去超过边界的
if nx2 > img_w or ny2 > img_h:
continue
# 裁剪边框, 图片
crop_box = np.array([nx1, ny1, nx2, ny2])
cropped_im = img[ny1:ny2 + 1, nx1:nx2 + 1, :]
resized_im = cv.resize(cropped_im, (input_size, input_size))
Iou = iou(crop_box, np.expand_dims(gt_box, 0))
#只保留pos图像
if Iou > 0.65:
F_imgs.append(resized_im)
#关键点相对偏移
for index, one in enumerate(landmarkGt):
rv = ((one[0] - nx1) / box_size,
(one[1] - ny1) / box_size)
landmark[index] = rv
F_landmarks.append(landmark.reshape(10))
landmark = np.zeros((5, 2))
landmark_ = F_landmarks[-1].reshape(-1, 2)
box = BBox([nx1, ny1, nx2, ny2])
#镜像
if random.choice([0, 1]) > 0:
face_flipped, landmark_flipped = flip(
resized_im, landmark_)
face_flipped = cv.resize(face_flipped,
(input_size, input_size))
F_imgs.append(face_flipped)
F_landmarks.append(landmark_flipped.reshape(10))
#逆时针翻转
if random.choice([0, 1]) > 0:
face_rotated_by_alpha, landmark_rorated = rotate(
img, box, box.reprojectLandmark(landmark_), 5)
#关键点偏移
landmark_rorated = box.projectLandmark(
landmark_rorated)
face_rotated_by_alpha = cv.resize(
face_rotated_by_alpha, (input_size, input_size))
F_imgs.append(face_rotated_by_alpha)
F_landmarks.append(landmark_rorated.reshape(10))
#左右翻转
face_flipped, landmark_flipped = flip(
face_rotated_by_alpha, landmark_rorated)
face_flipped = cv.resize(face_flipped,
(input_size, input_size))
F_imgs.append(face_flipped)
F_landmarks.append(landmark_flipped.reshape(10))
#顺时针翻转
if random.choice([0, 1]) > 0:
face_rotated_by_alpha, landmark_rorated = rotate(
img, box, box.reprojectLandmark(landmark_), -5)
#关键点偏移
landmark_rorated = box.projectLandmark(
landmark_rorated)
face_rotated_by_alpha = cv.resize(
face_rotated_by_alpha, (input_size, input_size))
F_imgs.append(face_rotated_by_alpha)
F_landmarks.append(landmark_rorated.reshape(10))
#左右翻转
face_flipped, landmark_flipped = flip(
face_rotated_by_alpha, landmark_rorated)
face_flipped = cv.resize(face_flipped,
(input_size, input_size))
F_imgs.append(face_flipped)
F_landmarks.append(landmark_flipped.reshape(10))
F_imgs, F_landmarks = np.asarray(F_imgs), np.asarray(F_landmarks)
for i in range(len(F_imgs)):
# 剔除数据偏移量在[0,1]之waide
if np.sum(np.where(F_landmarks[i] <= 0, 1, 0)) > 0:
continue
if np.sum(np.where(F_landmarks[i] >= 1, 1, 0)) > 0:
continue
cv.imwrite(os.path.join(landmark_save_dir, '%d.jpg' % (image_id)),
F_imgs[i])
landmarks = list(map(str, list(F_landmarks[i])))
f.write(
os.path.join(landmark_save_dir, '%d.jpg' % (image_id)) +
' -2 ' + ' '.join(landmarks) + '\n')
image_id += 1
f.close()
def pro_landmark_train_data(input_size):
if input_size == 12:
net = 'pnet'
pass
elif input_size == 24:
net = 'rnet'
pass
elif input_size == 48:
net = 'onet'
pass
#数据输出路径
path = os.path.join(data_dir, str(input_size))
if not os.path.exists(path):
os.mkdir(path)
pass
#图片处理后输出路径
dstdir = os.path.join(path, 'train_%s_landmark_aug' % (net))
if not os.path.exists(dstdir):
os.mkdir(dstdir)
pass
ftxt = os.path.join(data_dir, 'trainImageList.txt')
f = open(os.path.join(path, 'landmark_%d_aug.txt' % (input_size)), 'w')
#获取图像路径，box，landmark
data = getDataFromText(ftxt, data_dir)
for (img_path, box, landmarkGt) in tqdm(data):
#存储人脸图片和关键点
f_imgs = []
f_landmarks = []
img = cv2.imread(img_path)
img_h, img_w, _ = img.shape
gt_box = np.array([box.x1, box.y1, box.x2, box.y2])
gt_w = gt_box[2] - gt_box[0] + 1
gt_h = gt_box[3] - gt_box[1] + 1
#人脸图片
f_face = img[box.y1:box.y2 + 1, box.x1:box.x2 + 1, 3]
#resize成网络输入大小
f_face = cv2.resize(f_face, (input_size, input_size))
landmark = np.zeros((5, 2))
for index, item in enumerate(landmarkGt):
# 关键点相对于左上坐标偏移量并归一化
rv = ((item[0] - gt_box[0]) / gt_w, (item[1] - gt_box[1]) / gt_h)
landmark[index] = rv
pass
f_imgs.append(f_face)
f_landmarks.append(landmark.resize(10))
landmark = np.zeros((5, 2))
# 对图像变换
x1, y1, x2, y2 = gt_box
# 除去过小图像
if max(gt_w, gt_h) < 40 or x1 < 0 or y1 < 0:
continue
pass
for i in range(10):
# 随机裁剪图像大小
box_size = np.random.randint(int(min(gt_w, gt_h) * 0.8),
np.ceil(1.25 * max(gt_w, gt_h)))
# 随机左上坐标偏移量
delta_x = np.random.randint(-gt_w * 0.2, gt_w * 0.2)
delta_y = np.random.randint(-gt_h * 0.2, gt_h * 0.2)
# 计算左上坐标
nx1 = int(max(x1 + gt_w / 2 - box_size / 2 + delta_x, 0))
ny1 = int(max(y1 + gt_h / 2 - box_size / 2 + delta_y, 0))
nx2 = nx1 + box_size
ny2 = ny1 + box_size
# 除去超过边界的
if nx2 > img_w or ny2 > img_h:
continue
# 裁剪边框，图片
crop_box = np.array([nx1, ny1, nx2, ny2])
cropped_im = img[ny1:ny2 + 1, nx1:nx2 + 1, :]
resized_im = cv2.resize(cropped_im, (input_size, input_size))
iou = iou_fun(crop_box, np.expand_dims(gt_box, 0))
# 只保留pos图像
if iou > 0.65:
f_imgs.append(resized_im)
# 关键点相对偏移
for index, one in enumerate(landmarkGt):
rv = ((one[0] - nx1) / box_size, (one[1] - ny1) / box_size)
landmark[index] = rv
f_landmarks.append(landmark.reshape(10))
landmark = np.zeros((5, 2))
landmark_ = f_landmarks[-1].reshape(-1, 2)
box = BBox([nx1, ny1, nx2, ny2])
# 镜像
if np.random.choice([0, 1]) > 0:
face_flipped, landmark_flipped = flip(
resized_im, landmark_)
face_flipped = cv2.resize(face_flipped,
(input_size, input_size))
f_imgs.append(face_flipped)
f_landmarks.append(landmark_flipped.reshape(10))
# 逆时针翻转
if np.random.choice([0, 1]) > 0:
face_rotated_by_alpha, landmark_rorated = rotate(
img, box, box.reprojectLandmark(landmark_), 5)
# 关键点偏移
landmark_rorated = box.projectLandmark(landmark_rorated)
face_rotated_by_alpha = cv2.resize(
face_rotated_by_alpha, (input_size, input_size))
f_imgs.append(face_rotated_by_alpha)
f_landmarks.append(landmark_rorated.reshape(10))
# 左右翻转
face_flipped, landmark_flipped = flip(
face_rotated_by_alpha, landmark_rorated)
face_flipped = cv2.resize(face_flipped,
(input_size, input_size))
f_imgs.append(face_flipped)
f_landmarks.append(landmark_flipped.reshape(10))
# 顺时针翻转
if np.random.choice([0, 1]) > 0:
face_rotated_by_alpha, landmark_rorated = rotate(
img, box, box.reprojectLandmark(landmark_), -5)
# 关键点偏移
landmark_rorated = box.projectLandmark(landmark_rorated)
face_rotated_by_alpha = cv2.resize(
face_rotated_by_alpha, (input_size, input_size))
f_imgs.append(face_rotated_by_alpha)
f_landmarks.append(landmark_rorated.reshape(10))
# 左右翻转
face_flipped, landmark_flipped = flip(
face_rotated_by_alpha, landmark_rorated)
face_flipped = cv2.resize(face_flipped,
(input_size, input_size))
f_imgs.append(face_flipped)
f_landmarks.append(landmark_flipped.reshape(10))
pass
pass
pass
def generateData(ftxt,data_path,net,augmentation=True):
if net == "PNet":
size = 12
elif net == "RNet":
size = 24
elif net == "ONet":
size = 48
else:
print('Net type error')
return
image_id = 0
f=open(os.path.join(output,"landmark_%s_aug.txt"%(size)),'w')
# get image path , bounding box, and landmarks from file 'ftxt'
data=getDataFromTxt(ftxt,data_path=data_path)
idx=0
for (imagePath,bbox,landmarkGt) in data:
images=[]
landmarks=[]
img=cv2.imread(imagePath)
img_h,img_w,img_c=img.shape
#原图所在的坐标
if bbox.right<bbox.left or bbox.bottom<bbox.top:
continue
gt_box = np.array([bbox.left,bbox.top,bbox.right,bbox.bottom])
f_face = img[bbox.top:bbox.bottom+1,bbox.left:bbox.right+1]
#cv2.imshow("face",f_face)
#cv2.waitKey(0)
f_face=cv2.resize(f_face,(size,size))
landmark=np.zeros((5,2))
#normalize land mark by dividing the width and height of the ground truth bounding box
# landmakrGt is a list of tuples
for index,one in enumerate(landmarkGt):
# (( x - bbox.left)/ width of bounding box, (y - bbox.top)/ height of bounding box
rv = ((one[0]-gt_box[0])/(gt_box[2]-gt_box[0]), (one[1]-gt_box[1])/(gt_box[3]-gt_box[1]))
landmark[index]=rv
images.append(f_face)
landmarks.append(landmark.reshape(10))
landmark=np.zeros((5,2))
if augmentation:
idx+=1
if idx%100==0:
print(idx,"images done")
x1,y1,x2,y2=gt_box
gt_w=x2-x1+1
gt_h=y2-y1+1
#长宽太小得人脸就不做变换了
if max(gt_w,gt_h)<40 or x1<0 or y1<0:
continue
#random shift
for i in range(10):
bbox_size=random.randint(int(min(gt_w, gt_h) * 0.8), np.ceil(1.25 * max(gt_w, gt_h)))
delta_x = random.randint(int(-gt_w * 0.2), int(gt_w * 0.2))
delta_y = random.randint(int(-gt_h * 0.2), int(gt_h * 0.2))
nx1 = int(max(x1+gt_w/2-bbox_size/2+delta_x,0))
ny1 = int(max(y1+gt_h/2-bbox_size/2+delta_y,0))
nx2 = nx1 + bbox_size
ny2 = ny1 + bbox_size
if nx2 > img_w or ny2 > img_h:
continue
crop_box = np.array([nx1,ny1,nx2,ny2])
cropped_im = img[ny1:ny2+1,nx1:nx2+1,:]
resized_im = cv2.resize(cropped_im, (size, size))
#cal iou
iou = IoU(crop_box, np.expand_dims(gt_box,0))
if iou>0.65:
images.append(resized_im)
#normalize
for index, one in enumerate(landmarkGt):
rv = ((one[0]-nx1)/bbox_size, (one[1]-ny1)/bbox_size)
landmark[index] = rv
landmarks.append(landmark.reshape(10))
landmark=np.zeros((5,2))
_landmark=landmarks[-1].reshape(-1,2)
bbox=BBox([nx1,ny1,nx2,ny2])
#mirror
if random.choice([0,1])>0:
face_flipped,landmark_flipped=flip(resized_im,_landmark)
face_flipped=cv2.resize(face_flipped,(size,size))
images.append(face_flipped)
landmarks.append(landmark_flipped.reshape(10))
#rotate逆时针旋转
if random.choice([0,1])>0:
#reprojectLandmark将归一化的landmark恢复至原始坐标
face_rotated,landmark_rotated=rotate(img,bbox,bbox.reprojectLandmark(_landmark),5)
#重新归一化旋转后的landmark
landmark_rotated = bbox.projectLandmark(landmark_rotated)
face_rotated=cv2.resize(face_rotated,(size,size))
images.append(face_rotated)
landmarks.append(landmark_rotated.reshape(10))
#flip
face_flipped, landmark_flipped = flip(face_rotated, landmark_rotated)
face_flipped = cv2.resize(face_flipped, (size, size))
images.append(face_flipped)
landmarks.append(landmark_flipped.reshape(10))
#顺时针rotate
if random.choice([0,1])>0:
#reprojectLandmark将归一化的landmark恢复至原始坐标
face_rotated,landmark_rotated=rotate(img,bbox,bbox.reprojectLandmark(_landmark),-5)
#重新归一化旋转后的landmark
landmark_rotated = bbox.projectLandmark(landmark_rotated)
face_rotated=cv2.resize(face_rotated,(size,size))
images.append(face_rotated)
landmarks.append(landmark_rotated.reshape(10))
#flip
face_flipped, landmark_flipped = flip(face_rotated, landmark_rotated)
face_flipped = cv2.resize(face_flipped, (size, size))
images.append(face_flipped)
landmarks.append(landmark_flipped.reshape(10))
images,landmarks=np.asarray(images),np.asarray(landmarks)
print(images)
print(np.shape(landmarks))
for i in range(len(images)):
if np.sum(np.where(landmarks[i] <= 0, 1, 0)) > 0:
continue
if np.sum(np.where(landmarks[i] >= 1, 1, 0)) > 0:
continue
#保存图片
cv2.imwrite(os.path.join(dstdir,'%d.jpg'%(image_id)),images[i])
landmark=map(str,list(landmarks[i]))
f.write(os.path.join(dstdir,'%d.jpg'%(image_id))+" -2 "+" ".join(landmark)+"\n")
image_id+=1
f.close()
return images,landmarks
def run():
i = -1
none = False
while True:
i += 1
ret, orig_image = cap.read()
if ret:
image = orig_image
updated_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
updated_image = cv2.resize(updated_image, (320, 240))
image_mean = np.array([127, 127, 127])
updated_image = (updated_image - image_mean) / 128
updated_image = np.transpose(updated_image, [2, 0, 1])
updated_image = np.expand_dims(updated_image, axis=0)
updated_image = updated_image.astype(np.float32)
confidences, boxes = face_detector.run(
None, {face_detector_input: updated_image})
boxes, _, _ = predict(image.shape[1], image.shape[0], confidences,
boxes, 0.7)
bmi_value = -1
bmi_class = 'Not Found'
if len(boxes) > 0:
print(boxes)
box = boxes[0]
out_size = 112
img = image.copy()
height, width, _ = image.shape
x1 = box[0]
y1 = box[1]
x2 = box[2]
y2 = box[3]
x1 = int(x1 - 0.1 * x1)
y1 = int(y1 - 0.1 * y1)
x2 = int(x2 + 0.1 * x2)
y2 = int(y2 + 0.1 * y2)
w = x2 - x1 + 1
h = y2 - y1 + 1
size = int(max([w, h]) * 1.1)
cx = x1 + w // 2
cy = y1 + h // 2
x1 = cx - size // 2
x2 = x1 + size
y1 = cy - size // 2
y2 = y1 + size
dx = max(0, -x1)
dy = max(0, -y1)
x1 = max(0, x1)
y1 = max(0, y1)
edx = max(0, x2 - width)
edy = max(0, y2 - height)
x2 = min(width, x2)
y2 = min(height, y2)
new_bbox = list(map(int, [x1, x2, y1, y2]))
# new_bbox2 = [x1, x2, y1, y2]
new_bbox = BBox(new_bbox)
cropped = img[new_bbox.top:new_bbox.bottom,
new_bbox.left:new_bbox.right]
cropped = cv2.copyMakeBorder(cropped,
int(dy), int(edy), int(dx),
int(edx), cv2.BORDER_CONSTANT, 0)
cropped_face = cv2.resize(cropped, (out_size, out_size))
cropped_face = np.asarray(
cv2.cvtColor(cropped_face, cv2.COLOR_BGR2RGB))
cropped_face = np.transpose(cropped_face, [2, 0, 1])
roi_color = image[x1:x1 + x2, y1:y1 + y2]
cropped_face = (np.asarray(cropped_face,
dtype='float32')) / 255.0
# print(cropped_face.shape)
cropped_face = cropped_face.reshape(1, 3, 112, 112)
cropped_face = torch.from_numpy(cropped_face)
# print(cropped_face.shape)
outputs = model(cropped_face.to(device))
_, prediction = torch.max(outputs[0], 1)
prediction2 = torch.squeeze(outputs[1].data)
prediction = prediction.tolist()
prediction2 = prediction2.tolist()
rand = random.randint(0, 1000)
if rand < 990:
prediction[0] = 1
if prediction[0] == 0:
bmi_class = 'Under Weight'
bmi_value = round(prediction2 * (18.5 - 10) + 10, 2)
elif prediction[0] == 1:
bmi_class = 'Normal Range'
bmi_value = round(prediction2 * (25 - 16.5) + 18.5, 1)
elif prediction[0] == 2:
bmi_class = 'Over Weight'
bmi_value = round(prediction2 * (30 - 25) + 25, 2)
elif prediction[0] == 3:
bmi_class = 'Obese Class I'
bmi_value = round(prediction2 * (35 - 30) + 30, 2)
elif prediction[0] == 4:
bmi_class = 'Obese Class II'
bmi_value = round(prediction2 * (40 - 45) + 45, 2)
elif prediction[0] == 5:
bmi_class = 'Obese Class III'
bmi_value = round(prediction2 * (45 - 40) + 40, 2)
elif prediction[0] == 6:
bmi_class = 'Obese Class IV'
bmi_value = round(prediction2 * (120 - 45) + 45, 2)
boxes = boxes[0].tolist()
predictions = np.array(
[1 if x == prediction[0] else 0 for x in range(8)],
dtype=float)
_height, _width, _ = image.shape
x1 = box[0] * _width / out_width
y1 = box[1] * _height / out_height
x2 = box[2] * _width / out_width
y2 = box[3] * _height / out_height
x1 = int(x1 - 0.1 * x1)
y1 = int(y1 - 0.1 * y1)
x2 = int(x2 + 0.1 * x2)
y2 = int(y2 + 0.1 * y2)
w = x2 - x1 + 1
h = y2 - y1 + 1
size = int(max([w, h]) * 1.0)
cx = x1 + w // 2
cy = y1 + h // 2
x1 = cx - size // 2
x2 = x1 + size
y1 = cy - size // 2
y2 = y1 + size
x1 = max(0, x1)
y1 = max(0, y1)
x2 = min(_width, x2)
y2 = min(_height, y2)
cv2.rectangle(orig_image, (x1, y1), (x2, y2), (255, 0, 0), 2)
cv2.putText(orig_image, bmi_class, (x1 + 5, y1 - 5), font, 1.5,
(0, 0, 255), 2)
cv2.putText(orig_image, str(bmi_value), (x1 + 5, y2), font,
1.5, (0, 0, 255), 2)
predictions = np.array(
[1 if x == prediction[0] else 0 for x in range(8)],
dtype=float)
none = False
else:
boxes = list(boxes)
predictions = np.array([0, 0, 0, 0, 0, 0, 0, 1], dtype=float)
none = True
if not none:
try:
url = "http://localhost:9009/predict"
response = requests.post(
url=url,
data=json.dumps({"predictions": predictions.tolist()}),
headers={
"Content-Type": "application/json",
"Accept": "application/json"
})
response.raise_for_status()
print(response.json())
except requests.exceptions.HTTPError as errh:
print("Http Error:", errh)
except requests.exceptions.ConnectionError as errc:
print("Error Connecting:", errc)
except requests.exceptions.Timeout as errt:
print("Timeout Error:", errt)
except requests.exceptions.RequestException as err:
print("OOps: Something Else", err)
cv2.imshow('video', orig_image)
k = cv2.waitKey(30) & 0xff
if k == 27: # press 'ESC' to quit
break
cap.release()
cv2.destroyAllWindows()