def geneDataTxt(imgPath,landmarkREF,mode = 'test'):
imgTxt = 'dataset/data/testImageList.txt'
data = getDataFromTxt(imgTxt,False,False) #image_path,bbox
testData = defaultdict(lambda:dict(landmarks=[],patches=[],label=[]))
logger("generate 25 positive samples and 500 negative samples for per landmark")
testData = generateSamples(testData,data,landmarkREF)
for idx,name in types:
patches = np.asarray(testData[name]['patches'])
landmarks = np.asarray(testData[name]['landmarks'])
#label = np.asarray(testData[name]['label'])
patches = processImage(patches)
shuffle_in_unison_scary(patches,landmarks)
createDir('dataset/test/%s' % imgPath[-7:-4])
with h5py.File('dataset/test/%s/%s.h5' % (imgPath[-7:-4],name),'w') as h5:
h5['data'] = patches.astype(np.float32)
h5['landmark'] = landmarks.astype(np.float32)
#h5['label'] = label.astype(np.uint8)
with open('dataset/test/%s/%s.txt' % (imgPath[-7:-4],name),'w') as fd:
fd.write('dataset/test/%s/%s.h5'% (imgPath[-7:-4],name))
'''with open('dataset/test/%s.txt' % (name),'w') as fd:
def geneDataTxt(imgPath, landmarks):
imgTxt = 'dataset/data/testImageList.txt'
data = getDataFromTxt(imgTxt, False, False) #image_path,bbox
trainData = defaultdict(lambda: dict(landmarks=[], patches=[], label=[]))
logger(
"generate 25 positive samples and 500 negative samples for per landmark"
)
trainData = generateSamples(trainData, data, landmarks)
arr = []
for idx, name in types:
patches = np.asarray(trainData[name]['patches'])
landmarks = np.asarray(trainData[name]['landmarks'])
label = np.asarray(trainData[name]['label'])
arr1 = landmarks.reshape(1, -1)
arr.append(arr1)
patches = processImage(patches)
shuffle_in_unison_scary(patches, landmarks, label)
with open('test/%s.txt' % (name), 'w') as fd:
fd.append(landmarks.astype(np.float32))
fd.append(patches.astype(np.float32))
fd.append(label.astype(np.uint8))
def load_all_shoes():
conn = MongoClient(databaseUri)
db = conn.getter
shoes = []
labels = []
ids = []
docs = db.items.find(
{
"detail.images": {
"$elemMatch": {
"_id": {
"$exists": True
}
}
}
}, {
"detail": 1
}).limit(shoe_limit)
fs = []
for doc in docs:
if len(doc["detail"]["images"]) == 7:
for image in doc["detail"]["images"]:
if 'z' in image and image[
"z"] == 90 and 'y' in image and image["y"] == 0:
info = {
"f":
image_source + image["path"],
"cat":
doc["detail"]["categories"][
len(doc["detail"]["categories"]) - 1],
"image_id":
image["_id"]
}
try:
fs.append(info)
except:
print info
print doc["_id"]["_id"]
counter = 0
for i in fs:
# cat = doc["shoe"]["categories"][len(doc["shoe"]["categories"]) - 1]
# f = "/getter_data/images/" + str(image["_id"]) + contentTypeExtension[image["content-type"]]
print "Processing file [%d / %d] \r" % (counter, len(fs))
counter += 1
print i
shoes.append(processImage(i["f"], auto_crop, SZ))
labels.append(shoeCategory[i["cat"]])
ids.append(i["image_id"])
# except Exception, e:
# print 'action=processingFile,_id=%s,file=%s,error=%s' % (doc["_id"]["_id"], f, e)
# raise e
#
print "datasetSize=%d" % (len(shoes))
return np.array(shoes), np.array(labels), np.array(ids)
def getResult(img,bbox):
# F
F,EN,NM = getCNNs(level = 1)
f_bbox = bbox.subBBox(-0.05, 1.05, -0.05, 1.05)
f_face = img[f_bbox.top:f_bbox.bottom+1,f_bbox.left:f_bbox.right+1]
f_face = cv2.resize(f_face, (39, 39))
en_face = f_face[:31, :]
nm_face = f_face[8:, :]
f_face = f_face.reshape((1, 1, 39, 39))
f_face = processImage(f_face)
f = F.forward(f_face)
# EN
# en_bbox = bbox.subBBox(-0.05, 1.05, -0.04, 0.84)
# en_face = img[en_bbox.top:en_bbox.bottom+1,en_bbox.left:en_bbox.right+1]
en_face = cv2.resize(en_face, (31, 39)).reshape((1, 1, 31, 39))
#en_landmark = (landmarkGt[:3, :] - landmark_pre[:3,:]).reshape((6))
en_face = processImage(en_face)
en = EN.forward(en_face)
#EN_imgs.append(en_face)
#EN_landmarks.append(en_landmark)
# NM
# nm_bbox = bbox.subBBox(-0.05, 1.05, 0.18, 1.05)
# nm_face = img[nm_bbox.top:nm_bbox.bottom+1,nm_bbox.left:nm_bbox.right+1]
nm_face = cv2.resize(nm_face, (31, 39)).reshape((1, 1, 31, 39))
#nm_landmark = (landmarkGt[2:, :] - landmark_pre[2:,:]).reshape((6))
nm_face = processImage(nm_face)
nm = NM.forward(nm_face)
#NM_imgs.append(nm_face)
#NM_landmarks.append(nm_landmark)
landmark = np.zeros((5, 2))
landmark[0] = (f[0]+en[0]) / 2
landmark[1] = (f[1]+en[1]) / 2
landmark[2] = (f[2]+en[2]+nm[0]) / 3
landmark[3] = (f[3]+nm[1]) / 2
landmark[4] = (f[4]+nm[2]) / 2
return landmark
def load_all_shoes():
conn = MongoClient(databaseUri)
db = conn.getter
shoes = []
labels = []
ids = []
docs = db.items.find({
"detail.images": {
"$elemMatch" : {
"_id": {
"$exists": True
}
}
}
},
{
"detail": 1
}
).limit(shoe_limit)
fs = []
for doc in docs:
if len(doc["detail"]["images"]) == 7:
for image in doc["detail"]["images"]:
if 'z' in image and image["z"] == 90 and 'y' in image and image["y"] == 0:
info = {
"f": image_source + image["path"],
"cat": doc["detail"]["categories"][len(doc["detail"]["categories"]) - 1] ,
"image_id": image["_id"]
}
try:
fs.append(info)
except:
print info
print doc["_id"]["_id"]
counter = 0;
for i in fs:
# cat = doc["shoe"]["categories"][len(doc["shoe"]["categories"]) - 1]
# f = "/getter_data/images/" + str(image["_id"]) + contentTypeExtension[image["content-type"]]
print "Processing file [%d / %d] \r" % (counter, len(fs))
counter += 1
print i
shoes.append(processImage(i["f"], auto_crop, SZ))
labels.append(shoeCategory[i["cat"]])
ids.append(i["image_id"])
# except Exception, e:
# print 'action=processingFile,_id=%s,file=%s,error=%s' % (doc["_id"]["_id"], f, e)
# raise e
#
print "datasetSize=%d" % (len(shoes))
return np.array(shoes), np.array(labels), np.array(ids)
def NM(img, bbox):
"""
LEVEL-1, NM
img: gray image
bbox: bounding box of face
"""
bbox = bbox.expand(0.05)
face = img[bbox.top:bbox.bottom + 1, bbox.left:bbox.right + 1]
face = cv2.resize(face, (39, 39)).reshape((1, 1, 39, 39))
face = processImage(face)
F, EN, NM = getCNNs(level=1) # TODO more flexible load needed.
landmark = NM.forward(face[:, :, 8:, :])
return landmark
def NM(img, bbox):
"""
LEVEL-1, NM
img: gray image
bbox: bounding box of face
"""
bbox = bbox.expand(0.05)
face = img[bbox.top:bbox.bottom+1,bbox.left:bbox.right+1]
face = cv2.resize(face, (39, 39)).reshape((1, 1, 39, 39))
face = processImage(face)
F, EN, NM = getCNNs(level=1) # TODO more flexible load needed.
landmark = NM.forward(face[:, :, 8:, :])
return landmark
def generate(ftxt, mode, argument=False):
'''
第二阶段数据源制作
:param ftxt: 数据源文件位置和label
:param mode: 训练或测试
:param argument:
:return:
'''
data = getDataFromTxt(ftxt)
trainData = defaultdict(lambda: dict(patches=[], landmarks=[]))
for (imgPath, bbox, landmarkGt) in data:
img = cv2.imread(imgPath, cv2.IMREAD_GRAYSCALE)
assert (img is not None)
logger("process %s" % imgPath)
landmarkPs = randomShiftWithArgument(landmarkGt, 0.05)
if not argument:
landmarkPs = [landmarkPs[0]]
for landmarkP in landmarkPs:
for idx, name, padding in types:
patch, patch_bbox = getPatch(img, bbox, landmarkP[idx],
padding)
patch = cv2.resize(patch, (15, 15))
patch = patch.reshape((1, 15, 15))
trainData[name]['patches'].append(patch)
_ = patch_bbox.project(bbox.reproject(landmarkGt[idx]))
trainData[name]['landmarks'].append(_)
for idx, name, padding in types:
logger('writing training data of %s' % name)
patches = np.asarray(trainData[name]['patches'])
landmarks = np.asarray(trainData[name]['landmarks'])
patches = processImage(patches)
shuffle_in_unison_scary(patches, landmarks)
with h5py.File(
'/python/face_key_point/data_hdf5/train/2_%s/%s.h5' %
(name, mode), 'w') as h5:
h5['data'] = patches.astype(np.float32)
h5['landmark'] = landmarks.astype(np.float32)
with open(
'/python/face_key_point/data_hdf5/train/2_%s/%s.txt' %
(name, mode), 'w') as fd:
fd.write('/python/face_key_point/data_hdf5/train/2_%s/%s.h5' %
(name, mode))
def getResult(img,bbox):
# F
F,EN,NM = getCNNs(level = 1)
f_bbox = bbox.subBBox(-0.05, 1.05, -0.05, 1.05)
f_face = img[f_bbox.top:f_bbox.bottom+1,f_bbox.left:f_bbox.right+1]
f_face = cv2.resize(f_face, (39, 39))
en_face = f_face[:31, :]
nm_face = f_face[8:, :]
f_face = f_face.reshape((1, 1, 39, 39))
f_face = processImage(f_face)
f = F.forward(f_face)
return f
def generate(ftxt, mode, argument=False):
"""
Generate Training Data for LEVEL-3
mode = train or test
"""
data = getDataFromTxt(ftxt)
trainData = defaultdict(lambda: dict(patches=[], landmarks=[]))
for (imgPath, bbox, landmarkGt) in data:
img = cv2.imread(imgPath, cv2.CV_LOAD_IMAGE_GRAYSCALE)
assert (img is not None)
logger("process %s" % imgPath)
landmarkPs = randomShiftWithArgument(landmarkGt, 0.01)
if not argument:
landmarkPs = [landmarkPs[0]]
for landmarkP in landmarkPs:
for idx, name, padding in types:
patch, patch_bbox = getPatch(img, bbox, landmarkP[idx],
padding)
patch = cv2.resize(patch, (15, 15))
patch = patch.reshape((1, 15, 15))
trainData[name]['patches'].append(patch)
_ = patch_bbox.project(bbox.reproject(landmarkGt[idx]))
trainData[name]['landmarks'].append(_)
for idx, name, padding in types:
logger('writing training data of %s' % name)
patches = np.asarray(trainData[name]['patches'])
landmarks = np.asarray(trainData[name]['landmarks'])
patches = processImage(patches)
shuffle_in_unison_scary(patches, landmarks)
with h5py.File(
'/home/tyd/下载/deep_landmark/mydataset/mytrain/3_%s/%s.h5' %
(name, mode), 'w') as h5:
h5['data'] = patches.astype(np.float32)
h5['landmark'] = landmarks.astype(np.float32)
with open(
'/home/tyd/下载/deep_landmark/mydataset/mytrain/3_%s/%s.txt' %
(name, mode), 'w') as fd:
fd.write(
'/home/tyd/下载/deep_landmark/mydataset/mytrain/3_%s/%s.h5' %
(name, mode))
def generate_hdf5():
labelfile =readdata('../data/my_alige_landmark.txt')
F_imgs = []
F_landmarks = []
for i,l in enumerate(labelfile):
imgpath='../data/'+l[0]
img=cv2.imread(imgpath)
maxx=max(img.shape[0],img.shape[1])
img=sqrtimg(img)#把输入图片填充成正方形，因为我们要训练的图片的大小是正方形的图片255*255
img=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)#图片转为灰度图像
f_face=cv2.resize(img,(39,39))#把图片缩放成255＊255的图片
# F
plt.imshow(f_face,cmap='gray')
f_face = f_face.reshape((1, 39, 39))
f_landmark =np.asarray(l[1:],dtype='float')
F_imgs.append(f_face)
#归一化人脸特征点标签,因为上面height等于width，这里比较懒，直接简写
f_landmark=f_landmark/maxx #归一化到0～１之间
print f_landmark
F_landmarks.append(f_landmark)
F_imgs, F_landmarks = np.asarray(F_imgs), np.asarray(F_landmarks)
F_imgs = processImage(F_imgs)#图片预处理，包含均值归一化，方差归一化等
shuffle_in_unison_scary(F_imgs, F_landmarks)#打乱数据
#生成h5py格式
with h5py.File(os.getcwd()+ '/train_data.h5', 'w') as f:
f['data'] = F_imgs.astype(np.float32)
f['landmark'] = F_landmarks.astype(np.float32)
#因为caffe的输入h5py不是直接使用上面的数据，而是需要调用.txt格式的文件
with open(os.getcwd() + '/train.txt', 'w') as f:
f.write(os.getcwd() + '/train_data.h5\n')
print i
def generate(txt, mode, N):
"""
generate Training Data for LEVEL-2: patches around landmarks
mode: train or validate
"""
assert (mode == 'train')
#从txt文件中获取数据
data = getDataFromTxt(txt, True,
True) #return [(image_path,landmarks,bbox)]
#用defaultdict存储数据
trainData = defaultdict(lambda: dict(landmarks=[], patches=[], label=[]))
logger(
"generate 25 positive samples and 500 negative samples for per landmark"
)
#产生正负训练样本
trainData = generateSamples(trainData, data)
print 'All data:' + str(len(trainData))
#print trainData['L1']
#arr = []
for idx, name in types: #19个点
logger('writing training data of %s' % name)
patches = np.asarray(trainData[name]['patches'])
landmarks = np.asarray(trainData[name]['landmarks'])
label = np.asarray(trainData[name]['label'])
#arr1 = landmarks.reshape(1,-1)
#arr.append(arr1)
patches = processImage(patches)
shuffle_in_unison_scary(patches, landmarks, label)
#将训练数据保存为hdf5格式
with h5py.File('train/2_%s/%s.h5' % (name, mode), 'w') as h5:
h5['data'] = patches.astype(np.float32)
h5['landmark'] = landmarks.astype(np.float32)
h5['label'] = label.astype(np.uint8)
with open('train/2_%s/%s.txt' % (name, mode), 'w') as fd:
fd.write('train/2_%s/%s.h5' % (name, mode))
#暂时只产生一个点的数据
if idx == 1:
break
def generate(ftxt, mode, argument=False):
"""
Generate Training Data for LEVEL-3
mode = train or test
"""
data = getDataFromTxt(ftxt)
trainData = defaultdict(lambda: dict(patches=[], landmarks=[]))
for (imgPath, bbox, landmarkGt) in data:
img = cv2.imread(imgPath, cv2.CV_LOAD_IMAGE_GRAYSCALE)
assert(img is not None)
logger("process %s" % imgPath)
landmarkPs = randomShiftWithArgument(landmarkGt, 0.01)
if not argument:
landmarkPs = [landmarkPs[0]]
for landmarkP in landmarkPs:
for idx, name, padding in types:
patch, patch_bbox = getPatch(img, bbox, landmarkP[idx], padding)
patch = cv2.resize(patch, (15, 15))
patch = patch.reshape((1, 15, 15))
trainData[name]['patches'].append(patch)
_ = patch_bbox.project(bbox.reproject(landmarkGt[idx]))
trainData[name]['landmarks'].append(_)
for idx, name, padding in types:
logger('writing training data of %s'%name)
patches = np.asarray(trainData[name]['patches'])
landmarks = np.asarray(trainData[name]['landmarks'])
patches = processImage(patches)
shuffle_in_unison_scary(patches, landmarks)
with h5py.File('train/3_%s/%s.h5'%(name, mode), 'w') as h5:
h5['data'] = patches.astype(np.float32)
h5['landmark'] = landmarks.astype(np.float32)
with open('train/3_%s/%s.txt'%(name, mode), 'w') as fd:
fd.write('train/3_%s/%s.h5'%(name, mode))
def generate_hdf5(ftxt, output, fname, argument=False):
data = getDataFromTxt(ftxt)
F_imgs = []
F_landmarks = []
EN_imgs = []
EN_landmarks = []
NM_imgs = []
NM_landmarks = []
for (imgPath, bbox, landmarkGt) in data:
img = cv2.imread(imgPath, cv2.IMREAD_GRAYSCALE) #读取灰度图
assert (img is not None)
logger("process %s" % imgPath)
# F
f_bbox = bbox.subBBox(-0.05, 1.05, -0.05, 1.05)
f_face = img[f_bbox.top:f_bbox.bottom + 1,
f_bbox.left:f_bbox.right + 1] #截图
## data argument
if argument and np.random.rand() > -1:
### flip
face_flipped, landmark_flipped = flip(
f_face, landmarkGt) #图片翻转，数据增强，相应坐标也要发生变换
face_flipped = cv2.resize(face_flipped, (39, 39)) #固定图片大小
F_imgs.append(face_flipped.reshape((
1, 39,
39))) #caffe要求输入的格式 ['batch_size','channel','height','width']
F_landmarks.append(landmark_flipped.reshape(10)) #把10个关键坐标点值变成一维向量
### rotation
# if np.random.rand() > 0.5:
# face_rotated_by_alpha, landmark_rotated = rotate(img, f_bbox,
# bbox.reprojectLandmark(landmarkGt), 5)
# landmark_rotated = bbox.projectLandmark(landmark_rotated)
# face_rotated_by_alpha = cv2.resize(face_rotated_by_alpha, (39, 39))
# F_imgs.append(face_rotated_by_alpha.reshape((1, 39, 39)))
# F_landmarks.append(landmark_rotated.reshape(10))
# ### flip with rotation
# face_flipped, landmark_flipped = flip(face_rotated_by_alpha, landmark_rotated)
# face_flipped = cv2.resize(face_flipped, (39, 39))
# F_imgs.append(face_flipped.reshape((1, 39, 39)))
# F_landmarks.append(landmark_flipped.reshape(10))
# ### rotation
# if np.random.rand() > 0.5:
# face_rotated_by_alpha, landmark_rotated = rotate(img, f_bbox,
# bbox.reprojectLandmark(landmarkGt), -5)
# landmark_rotated = bbox.projectLandmark(landmark_rotated)
# face_rotated_by_alpha = cv2.resize(face_rotated_by_alpha, (39, 39))
# F_imgs.append(face_rotated_by_alpha.reshape((1, 39, 39)))
# F_landmarks.append(landmark_rotated.reshape(10))
# ### flip with rotation
# face_flipped, landmark_flipped = flip(face_rotated_by_alpha, landmark_rotated)
# face_flipped = cv2.resize(face_flipped, (39, 39))
# F_imgs.append(face_flipped.reshape((1, 39, 39)))
# F_landmarks.append(landmark_flipped.reshape(10))
f_face = cv2.resize(f_face, (39, 39))
en_face = f_face[:31, :]
nm_face = f_face[8:, :]
f_face = f_face.reshape((1, 39, 39))
f_landmark = landmarkGt.reshape((10))
F_imgs.append(f_face)
F_landmarks.append(f_landmark)
# EN
# en_bbox = bbox.subBBox(-0.05, 1.05, -0.04, 0.84)
# en_face = img[en_bbox.top:en_bbox.bottom+1,en_bbox.left:en_bbox.right+1]
## data argument
if argument and np.random.rand() > 0.5:
### flip
face_flipped, landmark_flipped = flip(en_face, landmarkGt)
face_flipped = cv2.resize(face_flipped, (31, 39)).reshape(
(1, 31, 39))
landmark_flipped = landmark_flipped[:3, :].reshape((6))
EN_imgs.append(face_flipped)
EN_landmarks.append(landmark_flipped)
en_face = cv2.resize(en_face, (31, 39)).reshape((1, 31, 39))
en_landmark = landmarkGt[:3, :].reshape((6))
EN_imgs.append(en_face)
EN_landmarks.append(en_landmark)
# NM
# nm_bbox = bbox.subBBox(-0.05, 1.05, 0.18, 1.05)
# nm_face = img[nm_bbox.top:nm_bbox.bottom+1,nm_bbox.left:nm_bbox.right+1]
## data argument
if argument and np.random.rand() > 0.5:
### flip
face_flipped, landmark_flipped = flip(nm_face, landmarkGt)
face_flipped = cv2.resize(face_flipped, (31, 39)).reshape(
(1, 31, 39))
landmark_flipped = landmark_flipped[2:, :].reshape((6))
NM_imgs.append(face_flipped)
NM_landmarks.append(landmark_flipped)
nm_face = cv2.resize(nm_face, (31, 39)).reshape((1, 31, 39))
nm_landmark = landmarkGt[2:, :].reshape((6))
NM_imgs.append(nm_face)
NM_landmarks.append(nm_landmark)
#imgs, landmarks = process_images(ftxt, output)
F_imgs, F_landmarks = np.asarray(F_imgs), np.asarray(F_landmarks)
EN_imgs, EN_landmarks = np.asarray(EN_imgs), np.asarray(EN_landmarks)
NM_imgs, NM_landmarks = np.asarray(NM_imgs), np.asarray(NM_landmarks)
F_imgs = processImage(F_imgs) #数据标准化
shuffle_in_unison_scary(F_imgs, F_landmarks) #随机打乱
EN_imgs = processImage(EN_imgs)
shuffle_in_unison_scary(EN_imgs, EN_landmarks)
NM_imgs = processImage(NM_imgs)
shuffle_in_unison_scary(NM_imgs, NM_landmarks)
# full face
base = join(OUTPUT, '1_F')
createDir(base)
output = join(base, fname) #D:.\deep_landmark\dataset\train\1_F\train.h5
logger("generate %s" % output)
with h5py.File(output, 'w') as h5:
h5['data'] = F_imgs.astype(np.float32)
h5['landmark'] = F_landmarks.astype(np.float32)
# eye and nose
base = join(OUTPUT, '1_EN')
createDir(base)
output = join(base, fname)
logger("generate %s" % output)
with h5py.File(output, 'w') as h5:
h5['data'] = EN_imgs.astype(np.float32)
h5['landmark'] = EN_landmarks.astype(np.float32)
# nose and mouth
base = join(OUTPUT, '1_NM')
createDir(base)
output = join(base, fname)
logger("generate %s" % output)
with h5py.File(output, 'w') as h5:
h5['data'] = NM_imgs.astype(np.float32)
h5['landmark'] = NM_landmarks.astype(np.float32)
def generate_hdf5(ftxt, output, fname, argument=False):
data = getDataFromTxt(ftxt)
F_imgs = []
F_landmarks = []
EN_imgs = []
EN_landmarks = []
NM_imgs = []
NM_landmarks = []
for (imgPath, bbox, landmarkGt) in data:
img = cv2.imread(imgPath, cv2.CV_LOAD_IMAGE_GRAYSCALE)
assert(img is not None)
logger("process %s" % imgPath)
# F
f_bbox = bbox.subBBox(-0.05, 1.05, -0.05, 1.05)
f_face = img[f_bbox.top:f_bbox.bottom+1,f_bbox.left:f_bbox.right+1]
## data argument
if argument and np.random.rand() > -1:
### flip
face_flipped, landmark_flipped = flip(f_face, landmarkGt)
face_flipped = cv2.resize(face_flipped, (39, 39))
F_imgs.append(face_flipped.reshape((1, 39, 39)))
F_landmarks.append(landmark_flipped.reshape(10))
### rotation
if np.random.rand() > 0.5:
face_rotated_by_alpha, landmark_rotated = rotate(img, f_bbox, \
bbox.reprojectLandmark(landmarkGt), 5)
landmark_rotated = bbox.projectLandmark(landmark_rotated)
face_rotated_by_alpha = cv2.resize(face_rotated_by_alpha, (39, 39))
F_imgs.append(face_rotated_by_alpha.reshape((1, 39, 39)))
F_landmarks.append(landmark_rotated.reshape(10))
### flip with rotation
face_flipped, landmark_flipped = flip(face_rotated_by_alpha, landmark_rotated)
face_flipped = cv2.resize(face_flipped, (39, 39))
F_imgs.append(face_flipped.reshape((1, 39, 39)))
F_landmarks.append(landmark_flipped.reshape(10))
### rotation
if np.random.rand() > 0.5:
face_rotated_by_alpha, landmark_rotated = rotate(img, f_bbox, \
bbox.reprojectLandmark(landmarkGt), -5)
landmark_rotated = bbox.projectLandmark(landmark_rotated)
face_rotated_by_alpha = cv2.resize(face_rotated_by_alpha, (39, 39))
F_imgs.append(face_rotated_by_alpha.reshape((1, 39, 39)))
F_landmarks.append(landmark_rotated.reshape(10))
### flip with rotation
face_flipped, landmark_flipped = flip(face_rotated_by_alpha, landmark_rotated)
face_flipped = cv2.resize(face_flipped, (39, 39))
F_imgs.append(face_flipped.reshape((1, 39, 39)))
F_landmarks.append(landmark_flipped.reshape(10))
f_face = cv2.resize(f_face, (39, 39))
en_face = f_face[:31, :]
nm_face = f_face[8:, :]
f_face = f_face.reshape((1, 39, 39))
f_landmark = landmarkGt.reshape((10))
F_imgs.append(f_face)
F_landmarks.append(f_landmark)
# EN
# en_bbox = bbox.subBBox(-0.05, 1.05, -0.04, 0.84)
# en_face = img[en_bbox.top:en_bbox.bottom+1,en_bbox.left:en_bbox.right+1]
## data argument
if argument and np.random.rand() > 0.5:
### flip
face_flipped, landmark_flipped = flip(en_face, landmarkGt)
face_flipped = cv2.resize(face_flipped, (31, 39)).reshape((1, 31, 39))
landmark_flipped = landmark_flipped[:3, :].reshape((6))
EN_imgs.append(face_flipped)
EN_landmarks.append(landmark_flipped)
en_face = cv2.resize(en_face, (31, 39)).reshape((1, 31, 39))
en_landmark = landmarkGt[:3, :].reshape((6))
EN_imgs.append(en_face)
EN_landmarks.append(en_landmark)
# NM
# nm_bbox = bbox.subBBox(-0.05, 1.05, 0.18, 1.05)
# nm_face = img[nm_bbox.top:nm_bbox.bottom+1,nm_bbox.left:nm_bbox.right+1]
## data argument
if argument and np.random.rand() > 0.5:
### flip
face_flipped, landmark_flipped = flip(nm_face, landmarkGt)
face_flipped = cv2.resize(face_flipped, (31, 39)).reshape((1, 31, 39))
landmark_flipped = landmark_flipped[2:, :].reshape((6))
NM_imgs.append(face_flipped)
NM_landmarks.append(landmark_flipped)
nm_face = cv2.resize(nm_face, (31, 39)).reshape((1, 31, 39))
nm_landmark = landmarkGt[2:, :].reshape((6))
NM_imgs.append(nm_face)
NM_landmarks.append(nm_landmark)
#imgs, landmarks = process_images(ftxt, output)
F_imgs, F_landmarks = np.asarray(F_imgs), np.asarray(F_landmarks)
EN_imgs, EN_landmarks = np.asarray(EN_imgs), np.asarray(EN_landmarks)
NM_imgs, NM_landmarks = np.asarray(NM_imgs),np.asarray(NM_landmarks)
F_imgs = processImage(F_imgs)
shuffle_in_unison_scary(F_imgs, F_landmarks)
EN_imgs = processImage(EN_imgs)
shuffle_in_unison_scary(EN_imgs, EN_landmarks)
NM_imgs = processImage(NM_imgs)
shuffle_in_unison_scary(NM_imgs, NM_landmarks)
# full face
base = join(OUTPUT, '1_F')
createDir(base)
output = join(base, fname)
logger("generate %s" % output)
with h5py.File(output, 'w') as h5:
h5['data'] = F_imgs.astype(np.float32)
h5['landmark'] = F_landmarks.astype(np.float32)
# eye and nose
base = join(OUTPUT, '1_EN')
createDir(base)
output = join(base, fname)
logger("generate %s" % output)
with h5py.File(output, 'w') as h5:
h5['data'] = EN_imgs.astype(np.float32)
h5['landmark'] = EN_landmarks.astype(np.float32)
# nose and mouth
base = join(OUTPUT, '1_NM')
createDir(base)
output = join(base, fname)
logger("generate %s" % output)
with h5py.File(output, 'w') as h5:
h5['data'] = NM_imgs.astype(np.float32)
h5['landmark'] = NM_landmarks.astype(np.float32)
def generate_hdf5(ftxt, output, fname, argument=False):
'''
生成hdf5
:param ftxt: 图片路径对应人脸框和左眼睛、右眼睛、鼻子、左嘴角、右嘴角对应坐标
:param output: hdf5存放位置
:param fname: 保存名称
:param argument:
:return:
'''
data = getDataFromTxt(ftxt)
# 全局图片
F_imgs = []
# 全局坐标
F_landmarks = []
# 只包含了eye nose的图片
EN_imgs = []
# 只包含了eye nose的坐标
EN_landmarks = []
# 只包含nose mouth的图片
NM_imgs = []
# 只包含了nose mouth的坐标
NM_landmarks = []
for (imgPath, bbox, landmarkGt) in data:
img = cv2.imread(imgPath, cv2.IMREAD_GRAYSCALE)
assert(img is not None)
logger("process %s" % imgPath)
# 人脸框有点小，扩大一点
f_bbox = bbox.subBBox(-0.05, 1.05, -0.05, 1.05)
# 人脸框里面的图像
f_face = img[int(f_bbox.top):int(f_bbox.bottom+1),int(f_bbox.left):int(f_bbox.right+1)]
## data argument
if argument and np.random.rand() > -1:
### flip 图片水平翻转，数据增强
face_flipped, landmark_flipped = flip(f_face, landmarkGt)
face_flipped = cv2.resize(face_flipped, (39, 39))
# 1表示通道
F_imgs.append(face_flipped.reshape((1, 39, 39)))
# 5*2的关键点转换为一位数组
F_landmarks.append(landmark_flipped.reshape(10))
### rotation
if np.random.rand() > 0.5:
face_rotated_by_alpha, landmark_rotated = rotate(img, f_bbox, \
bbox.reprojectLandmark(landmarkGt), 5)
landmark_rotated = bbox.projectLandmark(landmark_rotated)
face_rotated_by_alpha = cv2.resize(face_rotated_by_alpha, (39, 39))
F_imgs.append(face_rotated_by_alpha.reshape((1, 39, 39)))
F_landmarks.append(landmark_rotated.reshape(10))
### flip with rotation
face_flipped, landmark_flipped = flip(face_rotated_by_alpha, landmark_rotated)
face_flipped = cv2.resize(face_flipped, (39, 39))
F_imgs.append(face_flipped.reshape((1, 39, 39)))
F_landmarks.append(landmark_flipped.reshape(10))
### rotation
if np.random.rand() > 0.5:
face_rotated_by_alpha, landmark_rotated = rotate(img, f_bbox, \
bbox.reprojectLandmark(landmarkGt), -5)
landmark_rotated = bbox.projectLandmark(landmark_rotated)
face_rotated_by_alpha = cv2.resize(face_rotated_by_alpha, (39, 39))
F_imgs.append(face_rotated_by_alpha.reshape((1, 39, 39)))
F_landmarks.append(landmark_rotated.reshape(10))
### flip with rotation
face_flipped, landmark_flipped = flip(face_rotated_by_alpha, landmark_rotated)
face_flipped = cv2.resize(face_flipped, (39, 39))
F_imgs.append(face_flipped.reshape((1, 39, 39)))
F_landmarks.append(landmark_flipped.reshape(10))
f_face = cv2.resize(f_face, (39, 39))
# 眼睛和鼻子
en_face = f_face[:31, :]
# 鼻子和嘴巴
nm_face = f_face[8:, :]
f_face = f_face.reshape((1, 39, 39))
f_landmark = landmarkGt.reshape((10))
F_imgs.append(f_face)
F_landmarks.append(f_landmark)
# EN
# en_bbox = bbox.subBBox(-0.05, 1.05, -0.04, 0.84)
# en_face = img[en_bbox.top:en_bbox.bottom+1,en_bbox.left:en_bbox.right+1]
## data argument
if argument and np.random.rand() > 0.5:
### flip
face_flipped, landmark_flipped = flip(en_face, landmarkGt)
face_flipped = cv2.resize(face_flipped, (31, 39)).reshape((1, 31, 39))
landmark_flipped = landmark_flipped[:3, :].reshape((6))
EN_imgs.append(face_flipped)
EN_landmarks.append(landmark_flipped)
en_face = cv2.resize(en_face, (31, 39)).reshape((1, 31, 39))
en_landmark = landmarkGt[:3, :].reshape((6))
EN_imgs.append(en_face)
EN_landmarks.append(en_landmark)
# NM
# nm_bbox = bbox.subBBox(-0.05, 1.05, 0.18, 1.05)
# nm_face = img[nm_bbox.top:nm_bbox.bottom+1,nm_bbox.left:nm_bbox.right+1]
## data argument
if argument and np.random.rand() > 0.5:
### flip
face_flipped, landmark_flipped = flip(nm_face, landmarkGt)
face_flipped = cv2.resize(face_flipped, (31, 39)).reshape((1, 31, 39))
landmark_flipped = landmark_flipped[2:, :].reshape((6))
NM_imgs.append(face_flipped)
NM_landmarks.append(landmark_flipped)
nm_face = cv2.resize(nm_face, (31, 39)).reshape((1, 31, 39))
nm_landmark = landmarkGt[2:, :].reshape((6))
NM_imgs.append(nm_face)
NM_landmarks.append(nm_landmark)
#imgs, landmarks = process_images(ftxt, output)
F_imgs, F_landmarks = np.asarray(F_imgs), np.asarray(F_landmarks)
EN_imgs, EN_landmarks = np.asarray(EN_imgs), np.asarray(EN_landmarks)
NM_imgs, NM_landmarks = np.asarray(NM_imgs),np.asarray(NM_landmarks)
F_imgs = processImage(F_imgs)
shuffle_in_unison_scary(F_imgs, F_landmarks)
EN_imgs = processImage(EN_imgs)
shuffle_in_unison_scary(EN_imgs, EN_landmarks)
NM_imgs = processImage(NM_imgs)
shuffle_in_unison_scary(NM_imgs, NM_landmarks)
# full face
base = join(OUTPUT, '1_F')
createDir(base)
output = join(base, fname)
logger("generate %s" % output)
with h5py.File(output, 'w') as h5:
h5['data'] = F_imgs.astype(np.float32)
h5['landmark'] = F_landmarks.astype(np.float32)
# eye and nose
base = join(OUTPUT, '1_EN')
createDir(base)
output = join(base, fname)
logger("generate %s" % output)
with h5py.File(output, 'w') as h5:
h5['data'] = EN_imgs.astype(np.float32)
h5['landmark'] = EN_landmarks.astype(np.float32)
# nose and mouth
base = join(OUTPUT, '1_NM')
createDir(base)
output = join(base, fname)
logger("generate %s" % output)
with h5py.File(output, 'w') as h5:
h5['data'] = NM_imgs.astype(np.float32)
h5['landmark'] = NM_landmarks.astype(np.float32)
def generate_hdf5(ftxt, output, fname, argument=False):
data = getDataFromTxt(ftxt)
F_imgs = []
F_landmarks = []
EN_imgs = []
EN_landmarks = []
NM_imgs = []
NM_landmarks = []
for (imgPath, bbox, landmarkGt) in data:
img = cv2.imread(imgPath, cv2.CV_LOAD_IMAGE_GRAYSCALE)
assert (img is not None)
logger("process %s" % imgPath)
# F
f_bbox = bbox.subBBox(-0.05, 1.05, -0.05, 1.05)
f_face = img[f_bbox.top:f_bbox.bottom + 1,
f_bbox.left:f_bbox.right + 1]
## data argument
if argument and np.random.rand() > -1:
### flip
face_flipped, landmark_flipped = flip(f_face, landmarkGt)
face_flipped = cv2.resize(face_flipped, (39, 39))
F_imgs.append(face_flipped.reshape((1, 39, 39)))
F_landmarks.append(landmark_flipped.reshape(10))
### rotation
if np.random.rand() > 0.5:
face_rotated_by_alpha, landmark_rotated = rotate(img, f_bbox, \
bbox.reprojectLandmark(landmarkGt), 5)
landmark_rotated = bbox.projectLandmark(landmark_rotated)
face_rotated_by_alpha = cv2.resize(face_rotated_by_alpha,
(39, 39))
F_imgs.append(face_rotated_by_alpha.reshape((1, 39, 39)))
F_landmarks.append(landmark_rotated.reshape(10))
### flip with rotation
face_flipped, landmark_flipped = flip(face_rotated_by_alpha,
landmark_rotated)
face_flipped = cv2.resize(face_flipped, (39, 39))
F_imgs.append(face_flipped.reshape((1, 39, 39)))
F_landmarks.append(landmark_flipped.reshape(10))
### rotation
if np.random.rand() > 0.5:
face_rotated_by_alpha, landmark_rotated = rotate(img, f_bbox, \
bbox.reprojectLandmark(landmarkGt), -5)
landmark_rotated = bbox.projectLandmark(landmark_rotated)
face_rotated_by_alpha = cv2.resize(face_rotated_by_alpha,
(39, 39))
F_imgs.append(face_rotated_by_alpha.reshape((1, 39, 39)))
F_landmarks.append(landmark_rotated.reshape(10))
### flip with rotation
face_flipped, landmark_flipped = flip(face_rotated_by_alpha,
landmark_rotated)
face_flipped = cv2.resize(face_flipped, (39, 39))
F_imgs.append(face_flipped.reshape((1, 39, 39)))
F_landmarks.append(landmark_flipped.reshape(10))
f_face = cv2.resize(f_face, (39, 39))
en_face = f_face[:31, :]
nm_face = f_face[8:, :]
f_face = f_face.reshape((1, 39, 39))
f_landmark = landmarkGt.reshape((10))
F_imgs.append(f_face)
F_landmarks.append(f_landmark)
# EN
# en_bbox = bbox.subBBox(-0.05, 1.05, -0.04, 0.84)
# en_face = img[en_bbox.top:en_bbox.bottom+1,en_bbox.left:en_bbox.right+1]
## data argument
if argument and np.random.rand() > 0.5:
### flip
face_flipped, landmark_flipped = flip(en_face, landmarkGt)
face_flipped = cv2.resize(face_flipped, (31, 39)).reshape(
(1, 31, 39))
landmark_flipped = landmark_flipped[:3, :].reshape((6))
EN_imgs.append(face_flipped)
EN_landmarks.append(landmark_flipped)
en_face = cv2.resize(en_face, (31, 39)).reshape((1, 31, 39))
en_landmark = landmarkGt[:3, :].reshape((6))
EN_imgs.append(en_face)
EN_landmarks.append(en_landmark)
# NM
# nm_bbox = bbox.subBBox(-0.05, 1.05, 0.18, 1.05)
# nm_face = img[nm_bbox.top:nm_bbox.bottom+1,nm_bbox.left:nm_bbox.right+1]
## data argument
if argument and np.random.rand() > 0.5:
### flip
face_flipped, landmark_flipped = flip(nm_face, landmarkGt)
face_flipped = cv2.resize(face_flipped, (31, 39)).reshape(
(1, 31, 39))
landmark_flipped = landmark_flipped[2:, :].reshape((6))
NM_imgs.append(face_flipped)
NM_landmarks.append(landmark_flipped)
nm_face = cv2.resize(nm_face, (31, 39)).reshape((1, 31, 39))
nm_landmark = landmarkGt[2:, :].reshape((6))
NM_imgs.append(nm_face)
NM_landmarks.append(nm_landmark)
#imgs, landmarks = process_images(ftxt, output)
F_imgs, F_landmarks = np.asarray(F_imgs), np.asarray(F_landmarks)
EN_imgs, EN_landmarks = np.asarray(EN_imgs), np.asarray(EN_landmarks)
NM_imgs, NM_landmarks = np.asarray(NM_imgs), np.asarray(NM_landmarks)
F_imgs = processImage(F_imgs)
shuffle_in_unison_scary(F_imgs, F_landmarks)
EN_imgs = processImage(EN_imgs)
shuffle_in_unison_scary(EN_imgs, EN_landmarks)
NM_imgs = processImage(NM_imgs)
shuffle_in_unison_scary(NM_imgs, NM_landmarks)
# full face
base = join(OUTPUT, '1_F')
createDir(base)
output = join(base, fname)
logger("generate %s" % output)
with h5py.File(output, 'w') as h5:
h5['data'] = F_imgs.astype(np.float32)
h5['landmark'] = F_landmarks.astype(np.float32)
# eye and nose
base = join(OUTPUT, '1_EN')
createDir(base)
output = join(base, fname)
logger("generate %s" % output)
with h5py.File(output, 'w') as h5:
h5['data'] = EN_imgs.astype(np.float32)
h5['landmark'] = EN_landmarks.astype(np.float32)
# nose and mouth
base = join(OUTPUT, '1_NM')
createDir(base)
output = join(base, fname)
logger("generate %s" % output)
with h5py.File(output, 'w') as h5:
h5['data'] = NM_imgs.astype(np.float32)
h5['landmark'] = NM_landmarks.astype(np.float32)