def usfaces_df(queue):
"""Preprocess and augment US Face Database faces to data/. Returns pandas dataframe"""
usfaces_df = pd.read_excel("Full Attribute Scores/demographic & others labels/demographic-others-labels.xlsx")
usfaces_df = usfaces_df[["Filename", "Attractive"]]
usfaces_df = usfaces_df.drop_duplicates(["Filename"])
for face in usfaces_df["Filename"]:
base = os.path.splitext(face)[0]
try:
preprocess.resize("10k US Adult Faces Database/Face Images/{0}".format(face), "data/{0}".format(face))
preprocess.hflip("data/{0}".format(face), "data/{0}-F.jpg".format(base))
preprocess.add_noise("data/{0}".format(face), "data/{0}-N.jpg".format(base))
except:
usfaces_df = usfaces_df[usfaces_df.Filename != face]
flipped_df = usfaces_df.copy()
noisy_df = usfaces_df.copy()
flipped_df["Filename"] = flipped_df["Filename"].str[:-4] + "-F.jpg"
noisy_df["Filename"] = noisy_df["Filename"].str[:-4] + "-N.jpg"
df = pd.concat([usfaces_df, flipped_df, noisy_df], ignore_index=True)
df.columns = ["Face", "Rating"]
df["Rating"] *= 10.0 / 5.0
queue.put(df)
def scutfbp_df(queue):
"""Preprocess and augment SCUT-FBP faces to data/. Returns pandas dataframe"""
scutfbp_df = pd.read_excel("Rating_Collection/Attractiveness label.xlsx")
# Convert type of #Image column to str
scutfbp_df["#Image"] = scutfbp_df["#Image"].astype(str)
# Drop column Standard Deviation
scutfbp_df = scutfbp_df.drop("Standard Deviation", 1)
for face in os.listdir("Data_Collection"):
if face.endswith(".jpg"):
base = os.path.splitext(face)[0]
# Regex to find numbers at end of string
img_num = re.match(".*?([0-9]+)$", base).group(1)
try:
preprocess.resize("Data_Collection/{0}".format(face), "data/{0}".format(face))
preprocess.hflip("data/{0}".format(face), "data/{0}-F.jpg".format(base))
preprocess.add_noise("data/{0}".format(face), "data/{0}-N.jpg".format(base))
except:
scutfbp_df = scutfbp_df[getattr(scutfbp_df, "#Image") != img_num]
flipped_df = scutfbp_df.copy()
noisy_df = scutfbp_df.copy()
flipped_df["#Image"] = "SCUT-FBP-" + flipped_df["#Image"] + "-F.jpg"
noisy_df["#Image"] = "SCUT-FBP-" + noisy_df["#Image"] + "-N.jpg"
scutfbp_df["#Image"] = "SCUT-FBP-" + scutfbp_df["#Image"] + ".jpg"
df = pd.concat([scutfbp_df, flipped_df, noisy_df], ignore_index=True)
# Rename #Image -> Face and Attractiveness label -> Rating
df.columns = ["Face", "Rating"]
# Convert from 5 point scale to 10 point scale
df["Rating"] *= 10.0 / 5.0
queue.put(df)
def models_df(queue):
"""Preprocess and augment models.com faces to data/. Returns pandas dataframe"""
imgs = []
for i in range(1, 216):
url = "http://models.com/newfaces/page/{0}".format(i)
page = urllib2.urlopen(url).read()
soup = BeautifulSoup(page, "lxml")
for tag in soup.findAll("img", {"class": "attachment-square"}):
src = "http:{0}".format(tag["src"])
base = uuid.uuid4().hex
filename = base + ".jpg"
urllib.urlretrieve(src, "data/{0}".format(filename))
try:
preprocess.resize("data/{0}".format(filename), "data/{0}".format(filename))
preprocess.hflip("data/{0}".format(filename), "data/{0}-F.jpg".format(base))
preprocess.add_noise("data/{0}".format(filename), "data/{0}-N.jpg".format(base))
except:
os.remove("data/{0}".format(filename))
continue
imgs.append({"Face": filename, "Rating": 10})
imgs.append({"Face": "{0}-F.jpg".format(base), "Rating": 10})
imgs.append({"Face": "{0}-N.jpg".format(base), "Rating": 10})
df = pd.DataFrame(imgs)
queue.put(df)
def chicago_df(queue):
"""Preprocess and augment Chicago faces to data/. Returns pandas dataframe"""
chicago_df = pd.read_excel("CFD Version 2.0/CFD 2.0 Norming Data and Codebook.xlsx", skiprows=4)
chicago_df = chicago_df[["Target", "Attractive"]]
for dir in os.listdir("CFD Version 2.0/CFD 2.0 Images"):
if dir == ".DS_Store":
continue
for face in os.listdir("CFD Version 2.0/CFD 2.0 Images/{0}".format(dir)):
# Neutral faces
if face.endswith("N.jpg"):
# Is one face detected
try:
preprocess.resize(
"CFD Version 2.0/CFD 2.0 Images/{0}/{1}".format(dir, face), "data/{0}.jpg".format(dir)
)
preprocess.hflip("data/{0}.jpg".format(dir), "data/{0}-F.jpg".format(dir))
preprocess.add_noise("data/{0}.jpg".format(dir), "data/{0}-N.jpg".format(dir))
except:
chicago_df = chicago_df[chicago_df.Target != dir]
flipped_df = chicago_df.copy()
noisy_df = chicago_df.copy()
flipped_df["Target"] = flipped_df["Target"] + "-F.jpg"
noisy_df["Target"] = noisy_df["Target"] + "-N.jpg"
chicago_df["Target"] = chicago_df["Target"] + ".jpg"
df = pd.concat([chicago_df, flipped_df, noisy_df], ignore_index=True)
# Rename Target -> Face and Attractive -> Rating
df.columns = ["Face", "Rating"]
# Convert from 7 point scale to 10 point scale
df["Rating"] *= 10.0 / 7.0
queue.put(df)
def eccv_df(queue):
"""Preprocess and augment Gray et al. dataset to data/. Returns pandas dataframe"""
root = ET.parse("eccv2010_beauty_data/hotornot_face_all.xml").getroot()
childs = []
for child in root:
filename = os.path.split(child.attrib["filename"])[-1]
base = os.path.splitext(filename)[0]
try:
preprocess.resize(
"eccv2010_beauty_data/{0}".format(child.attrib["filename"]),
"data/{0}".format(base + ".jpg"),
crop=False,
)
preprocess.hflip("data/{0}".format(filename), "data/{0}-F.jpg".format(base))
preprocess.add_noise("data/{0}".format(filename), "data/{0}-N.jpg".format(base))
except:
continue
childs.append([base + "-F.jpg", float(child.attrib["score"])])
childs.append([base + "-N.jpg", float(child.attrib["score"])])
childs.append([base + ".jpg", float(child.attrib["score"])])
df = pd.DataFrame(childs, columns=["Face", "Rating"])
df["Rating"] += 4
df["Rating"] *= 10.0 / 8.0
queue.put(df)
