first commit

.gitignore

@@ -0,0 +1,105 @@
+### https://raw.github.com/github/gitignore/f57304e9762876ae4c9b02867ed0cb887316387e/Python.gitignore
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+env/
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*,cover
+.hypothesis/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# pyenv
+.python-version
+
+# celery beat schedule file
+celerybeat-schedule
+
+# SageMath parsed files
+*.sage.py
+
+# dotenv
+.env
+
+# virtualenv
+.venv
+venv/
+ENV/
+
+# Spyder project settings
+.spyderproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+/.idea/
+
+/data/
+/models/
+/checkpoints/

README.md

@@ -0,0 +1,36 @@
+# Age and Gender Estimation
+This is a Keras implementation of a CNN network for estimating age and gender from a face image.
+In training, [the IMDB-WIKI dataset](https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/) is used.
+
+## Dependencies
+- Python3.5+
+- Keras
+- scipy, numpy, Pandas, tqdm
+- OpenCV3
+
+## Usage
+
+Download the dataset. The dataset is downloaded and extracted to the `data` directory.
+
+```sh
+./download.sh
+```
+
+Filter out noise data and serialize images and labels for training into `.mat` file.
+Please check `check_dataset.ipynb` for the details of the dataset.
+```sh
+python create_db.py --output data/imdb_db.mat --db imdb --img_size 64
+```
+
+Train the network using the training data created above.
+
+```sh
+python3 train.py --input data/imdb_db.mat
+```
+
+## Network architecture
+In [the original paper](https://www.vision.ee.ethz.ch/en/publications/papers/articles/eth_biwi_01299.pdf), the pretrained VGG network is adopted.
+Here the Wide Residual Network (WideResNet) is trained from scratch.
+I modified the @asmith26's implementation of the WideResNet; two classification layers (for age and gender estimation) are added on the top of the WideResNet.
+Note that age and gender are estimated independently using different two CNNs.
+

create_db.py

@@ -0,0 +1,62 @@
+import numpy as np
+import cv2
+import scipy.io
+import argparse
+from tqdm import tqdm
+from utils import get_meta
+
+
+def get_args():
+    parser = argparse.ArgumentParser(description="This script automatically blur input video.")
+    parser.add_argument("--output", "-o", type=str, required=True,
+                        help="path to output database mat file")
+    parser.add_argument("--db", type=str, default="wiki",
+                        help="dataset; wiki or imdb")
+    parser.add_argument("--img_size", type=int, default=32,
+                        help="output image size")
+    parser.add_argument("--min_score", type=float, default=1.0,
+                        help="minimum face_score")
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+    output_path = args.output
+    db = args.db
+    img_size = args.img_size
+    min_score = args.min_score
+
+    root_path = "data/{}_crop/".format(db)
+    mat_path = root_path + "{}.mat".format(db)
+    full_path, dob, gender, photo_taken, face_score, second_face_score, age = get_meta(mat_path, db)
+
+    out_genders = []
+    out_ages = []
+    out_imgs = []
+
+    for i in tqdm(range(len(face_score))):
+        if face_score[i] < min_score:
+            continue
+
+        if (~np.isnan(second_face_score[i])) and second_face_score[i] > 0.0:
+            continue
+
+        if ~(0 <= age[i] <= 100):
+            continue
+
+        if np.isnan(gender[i]):
+            continue
+
+        out_genders.append(int(gender[i]))
+        out_ages.append(age[i])
+        img = cv2.imread(root_path + str(full_path[i][0]))
+        out_imgs.append(cv2.resize(img, (img_size, img_size)))
+
+    output = {"image": np.array(out_imgs), "gender": np.array(out_genders), "age": np.array(out_ages),
+              "db": db, "img_size": img_size, "min_score": min_score}
+    scipy.io.savemat(output_path, output)
+
+
+if __name__ == '__main__':
+    main()

download.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+mkdir -p data
+cd data
+
+if [ ! -f imdb_crop.tar ]; then
+    wget https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/static/imdb_crop.tar
+fi
+
+if [ ! -d imdb_crop ]; then
+    tar xf imdb_crop.tar
+fi
+
+if [ ! -f wiki_crop.tar ]; then
+    wget https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/static/wiki_crop.tar
+fi
+
+if [ ! -d wiki_crop ]; then
+    tar xf wiki_crop.tar
+fi

train.py

@@ -0,0 +1,94 @@
+import pandas as pd
+import logging
+import argparse
+import os
+from keras.callbacks import LearningRateScheduler, ModelCheckpoint
+from keras.optimizers import SGD
+from keras.utils import np_utils
+from wide_resnet import WideResNet
+from utils import mk_dir, load_data
+
+logging.basicConfig(level=logging.DEBUG)
+
+
+class Schedule:
+    def __init__(self, nb_epochs):
+        self.epochs = nb_epochs
+
+    def __call__(self, epoch_idx):
+        if epoch_idx < self.epochs * 0.25:
+            return 0.1
+        elif epoch_idx < self.epochs * 0.5:
+            return 0.02
+        elif epoch_idx < self.epochs * 0.75:
+            return 0.004
+        return 0.0008
+
+
+def get_args():
+    parser = argparse.ArgumentParser(description="This script trains the CNN model for age and gender estimation.")
+    parser.add_argument("--input", "-i", type=str, required=True,
+                        help="path to input database mat file")
+    parser.add_argument("--batch_size", type=int, default=32,
+                        help="batch size")
+    parser.add_argument("--nb_epochs", type=int, default=30,
+                        help="number of epochs")
+    parser.add_argument("--depth", type=int, default=16,
+                        help="depth of network")
+    parser.add_argument("--width", type=int, default=8,
+                        help="width of network")
+    parser.add_argument("--validation_split", type=float, default=0.1,
+                        help="validation split ratio")
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+    input_path = args.input
+    batch_size = args.batch_size
+    nb_epochs = args.nb_epochs
+    depth = args.depth
+    k = args.width
+    validation_split = args.validation_split
+
+    logging.debug("Loading data...")
+    image, gender, age, _, image_size, _ = load_data(input_path)
+    X_data = image
+    y_data_g = np_utils.to_categorical(gender, 2)
+    y_data_a = np_utils.to_categorical(age, 101)
+
+    model = WideResNet(image_size, depth=depth, k=k)()
+    sgd = SGD(lr=0.1, momentum=0.9, nesterov=True)
+    model.compile(optimizer=sgd, loss=["categorical_crossentropy", "categorical_crossentropy"],
+                  metrics=['accuracy'])
+
+    logging.debug("Model summary...")
+    model.count_params()
+    model.summary()
+
+    logging.debug("Saving model...")
+    mk_dir("models")
+    with open(os.path.join("models", "WRN_{}_{}.json".format(depth, k)), "w") as f:
+        f.write(model.to_json())
+
+    mk_dir("checkpoints")
+    callbacks = [LearningRateScheduler(schedule=Schedule(nb_epochs)),
+                 ModelCheckpoint("checkpoints/weights.{epoch:02d}-{val_loss:.2f}.hdf5",
+                                 monitor="val_loss",
+                                 verbose=1,
+                                 save_best_only=True,
+                                 mode="auto")
+                 ]
+
+    logging.debug("Running training...")
+    hist = model.fit(X_data, [y_data_g, y_data_a], batch_size=batch_size, nb_epoch=nb_epochs, callbacks=callbacks,
+                     validation_split=validation_split)
+
+    logging.debug("Saving weights...")
+    model.save_weights(os.path.join("models", "WRN_{}_{}.h5".format(depth, k)), overwrite=True)
+    pd.DataFrame(hist.history).to_hdf(os.path.join("models", "history_{}_{}.h5".format(depth, k)), "history")
+
+
+if __name__ == '__main__':
+    main()

utils.py

@@ -0,0 +1,39 @@
+from scipy.io import loadmat
+from datetime import datetime
+import os
+
+
+def calc_age(taken, dob):
+    birth = datetime.fromordinal(max(int(dob) - 366, 1))
+
+    # assume the photo was taken in the middle of the year
+    if birth.month < 7:
+        return taken - birth.year
+    else:
+        return taken - birth.year - 1
+
+
+def get_meta(mat_path, db):
+    meta = loadmat(mat_path)
+    full_path = meta[db][0, 0]["full_path"][0]
+    dob = meta[db][0, 0]["dob"][0]  # Matlab serial date number
+    gender = meta[db][0, 0]["gender"][0]
+    photo_taken = meta[db][0, 0]["photo_taken"][0]  # year
+    face_score = meta[db][0, 0]["face_score"][0]
+    second_face_score = meta[db][0, 0]["second_face_score"][0]
+    age = [calc_age(photo_taken[i], dob[i]) for i in range(len(dob))]
+
+    return full_path, dob, gender, photo_taken, face_score, second_face_score, age
+
+
+def load_data(mat_path):
+    d = loadmat(mat_path)
+
+    return d["image"], d["gender"][0], d["age"][0], d["db"][0], d["img_size"][0, 0], d["min_score"][0, 0]
+
+
+def mk_dir(dir):
+    try:
+        os.mkdir( dir )
+    except OSError:
+        pass