def test_openface():
# For OpenFace data file
filename = os.path.join(get_test_data_path(), "OpenFace_Test.csv")
openface = Fex(read_openface(filename), sampling_freq=30)
# Test KeyError
with pytest.raises(KeyError):
Fex(read_openface(filename, features=["NotHere"]), sampling_freq=30)
# Test length
assert len(openface) == 100
# Test loading from filename
openface = Fex(filename=filename, sampling_freq=30, detector="OpenFace")
openface = openface.read_file()
# Test length?
assert len(openface) == 100
# Test landmark methods
assert openface.landmark().shape[1] == 136
assert openface.iloc[0].landmark().shape[0] == 136
assert openface.landmark_x().shape[1] == openface.landmark_y().shape[1]
assert (openface.iloc[0].landmark_x().shape[0] ==
openface.iloc[0].landmark_y().shape[0])
# Test PSPI calculation b/c diff from facet
assert len(openface.calc_pspi()) == len(openface)
def test_openface():
# For OpenFace data file
filename = join(get_test_data_path(), 'OpenFace_Test.csv')
openface = Fex(read_openface(filename), sampling_freq=30)
# Test KeyError
with pytest.raises(KeyError):
Fex(read_openface(filename, features=['NotHere']), sampling_freq=30)
# Test length
assert len(openface) == 100
# Test loading from filename
openface = Openface(filename=filename, sampling_freq=30)
openface.read_file()
# Test length?
assert len(openface) == 100
# Test PSPI calculation b/c diff from facet
assert len(openface.calc_pspi()) == len(openface)
# Test if a method returns subclass.
openface = openface.downsample(target=10, target_type='hz')
assert isinstance(openface, Openface)
def test_stats():
filename = os.path.join(get_test_data_path(), "OpenFace_Test.csv")
openface = Fex(filename=filename, sampling_freq=30, detector="OpenFace")
openface = openface.read_file()
aus = openface.aus()
aus.sessions = range(len(aus))
y = aus[[i for i in aus.columns if "_r" in i]]
X = pd.DataFrame(aus.sessions)
b, t, p, df, res = aus.regress(X, y, mode="ols", fit_intercept=True)
assert b.shape == (2, 17)
assert res.mean().mean() < 1
clf = openface.predict(X=["AU02_c"], y="AU04_c")
assert clf.coef_ < 0
clf = openface.predict(X=openface[["AU02_c"]], y=openface["AU04_c"])
assert clf.coef_ < 0
t, p = openface[["AU02_c"]].ttest_1samp()
assert t > 0
a = openface.aus().assign(input="0")
b = openface.aus().apply(lambda x: x + np.random.rand(100)).assign(
input="1")
doubled = pd.concat([a, b])
doubled.sessions = doubled['input']
t, p = doubled.ttest_ind(col="AU12_r", sessions=("0", "1"))
assert (t < 0)
frame = np.concatenate([
np.array(range(int(len(doubled) / 2))),
np.array(range(int(len(doubled) / 2)))
])
assert (doubled.assign(frame=frame).isc(col="AU04_r").iloc[0, 0] == 1)
def test_affectiva():
filename = os.path.join(get_test_data_path(),
"sample_affectiva-api-app_output.json")
affdex = Fex(read_affectiva(filename),
sampling_freq=1,
detector="Affectiva")
assert affdex.shape[1] == 32
def detect_image(self, inputFname, outputFname=None):
"""Detects FEX from a video file.
Args:
inputFname (str, or list of str): Path to image file or a list of paths to image files.
outputFname (str, optional): Path to output file. Defaults to None.
Returns:
Fex: Prediction results dataframe if outputFname is None. Returns True if outputFname is specified.
"""
assert type(inputFname)==str or type(inputFname)==list, "inputFname must be a string path to image or list of image paths"
if type(inputFname)==str:
inputFname = [inputFname]
for inputF in inputFname:
if not os.path.exists(inputF):
raise FileNotFoundError(f"File {inputF} not found.")
self.info['inputFname'] = inputFname
init_df = pd.DataFrame(columns=self["output_columns"])
if outputFname:
init_df.to_csv(outputFname, index=False, header=True)
for inputF in inputFname:
frame = Image.open(inputF)
df = self.process_frame(np.array(frame))
if outputFname:
df.to_csv(outputFname, index=True, header=False, mode='a')
else:
init_df = pd.concat([init_df, df], axis=0)
if outputFname:
return True
else:
return Fex(init_df, filename = inputFname, au_columns = None, emotion_columns = FEAT_EMOTION_COLUMNS, facebox_columns = FEAT_FACEBOX_COLUMNS, landmark_columns = openface_2d_landmark_columns, time_columns = FACET_TIME_COLUMNS, detector="Feat")
def detect_image(self, inputFname, outputFname=None, verbose=False):
"""Detects FEX from an image file.
Args:
inputFname (str, or list of str): Path to image file or a list of paths to image files.
outputFname (str, optional): Path to output file. Defaults to None.
Rseturns:
Fex: Prediction results dataframe if outputFname is None. Returns True if outputFname is specified.
"""
assert (
type(inputFname) == str or type(inputFname) == list
), "inputFname must be a string path to image or list of image paths"
if type(inputFname) == str:
inputFname = [inputFname]
for inputF in inputFname:
if not os.path.exists(inputF):
raise FileNotFoundError(f"File {inputF} not found.")
self.info["inputFname"] = inputFname
init_df = pd.DataFrame(columns=self["output_columns"])
if outputFname:
init_df.to_csv(outputFname, index=False, header=True)
for inputF in inputFname:
if verbose:
print(f"processing {inputF}")
frame = cv2.imread(inputF)
df = self.process_frame(frame)
df["input"] = inputF
if outputFname:
df[init_df.columns].to_csv(outputFname,
index=False,
header=False,
mode="a")
else:
init_df = pd.concat([init_df, df[init_df.columns]], axis=0)
if outputFname:
return True
else:
return Fex(
init_df,
filename=inputFname,
au_columns=self['au_presence_columns'],
emotion_columns=FEAT_EMOTION_COLUMNS,
facebox_columns=FEAT_FACEBOX_COLUMNS,
landmark_columns=openface_2d_landmark_columns,
time_columns=FACET_TIME_COLUMNS,
detector="Feat",
)
def test_stats():
filename = os.path.join(get_test_data_path(), "OpenFace_Test.csv")
openface = Fex(filename=filename, sampling_freq=30, detector="OpenFace")
openface = openface.read_file()
aus = openface.aus()
aus.sessions = range(len(aus))
y = aus[[i for i in aus.columns if "_r" in i]]
X = pd.DataFrame(aus.sessions)
b, t, p, df, res = aus.regress(X, y, mode="ols", fit_intercept=True)
assert b.shape == (2, 17)
assert res.mean().mean() < 1
clf = openface.predict(X=["AU02_c"], y="AU04_c")
assert clf.coef_ < 0
clf = openface.predict(X=openface[["AU02_c"]], y=openface["AU04_c"])
assert clf.coef_ < 0
t, p = openface[["AU02_c"]].ttest()
assert t > 0
def test_feat():
filename = os.path.join(get_test_data_path(), "output.csv")
fex = Fex(filename=filename, detector="Feat")
fex = fex.read_file()
# test input
assert fex.input().values[0] == fex.iloc[0].input()
def test_fex():
with pytest.raises(Exception):
fex = Fex().read_feat()
with pytest.raises(Exception):
fex = Fex().read_facet()
with pytest.raises(Exception):
fex = Fex().read_openface()
with pytest.raises(Exception):
fex = Fex().read_affectiva()
# For iMotions-FACET data files
# test reading iMotions file < version 6
filename = os.path.join(get_test_data_path(), 'iMotions_Test_v5.txt')
dat = Fex(read_facet(filename), sampling_freq=30)
# test reading iMotions file > version 6
filename = os.path.join(get_test_data_path(), 'iMotions_Test_v6.txt')
df = read_facet(filename)
# Test slicing functions.
assert df.aus().shape == (519, 20)
assert df.emotions().shape == (519, 12)
assert df.facebox().shape == (519, 4)
assert df.time().shape[-1] == 4
assert df.design().shape[-1] == 4
sessions = np.array([[x] * 10
for x in range(1 + int(len(df) / 10))]).flatten()[:-1]
dat = Fex(df, sampling_freq=30, sessions=sessions)
dat = dat[[
'Joy', 'Anger', 'Surprise', 'Fear', 'Contempt', 'Disgust', 'Sadness',
'Confusion', 'Frustration', 'Neutral', 'Positive', 'Negative'
]]
# Test Session ValueError
with pytest.raises(ValueError):
Fex(df, sampling_freq=30, sessions=sessions[:10])
# Test KeyError
with pytest.raises(KeyError):
Fex(read_facet(filename, features=['NotHere']), sampling_freq=30)
# Test length
assert len(dat) == 519
# Test sessions generator
assert len(np.unique(dat.sessions)) == len([x for x in dat.itersessions()])
# Test metadata propagation
assert dat[['Joy']].sampling_freq == dat.sampling_freq
assert dat.iloc[:, 0].sampling_freq == dat.sampling_freq
# Test Downsample
assert len(dat.downsample(target=10)) == 52
# Test upsample
assert len(dat.upsample(target=60, target_type='hz')) == (len(dat) - 1) * 2
# Test interpolation
assert dat.interpolate(method='linear').isnull().sum(
)['Positive'] < dat.isnull().sum()['Positive']
dat = dat.interpolate(method='linear')
# Test distance
d = dat[['Positive']].distance()
assert isinstance(d, Adjacency)
assert d.square_shape()[0] == len(dat)
# Test Copy
assert isinstance(dat.copy(), Fex)
assert dat.copy().sampling_freq == dat.sampling_freq
# Test rectification
rectified = df.rectification()
assert df[df.au_columns].isna().sum()[0] < rectified[
rectified.au_columns].isna().sum()[0]
# Test pspi
assert len(df.calc_pspi()) == len(df)
# Test baseline
assert isinstance(dat.baseline(baseline='median'), Fex)
assert isinstance(dat.baseline(baseline='mean'), Fex)
assert isinstance(dat.baseline(baseline='begin'), Fex)
assert isinstance(dat.baseline(baseline=dat.mean()), Fex)
assert isinstance(dat.baseline(baseline='median', ignore_sessions=True),
Fex)
assert isinstance(dat.baseline(baseline='mean', ignore_sessions=True), Fex)
assert isinstance(dat.baseline(baseline=dat.mean(), ignore_sessions=True),
Fex)
assert isinstance(dat.baseline(baseline='median', normalize='pct'), Fex)
assert isinstance(dat.baseline(baseline='mean', normalize='pct'), Fex)
assert isinstance(dat.baseline(baseline=dat.mean(), normalize='pct'), Fex)
assert isinstance(
dat.baseline(baseline='median', ignore_sessions=True, normalize='pct'),
Fex)
assert isinstance(
dat.baseline(baseline='mean', ignore_sessions=True, normalize='pct'),
Fex)
assert isinstance(
dat.baseline(baseline=dat.mean(),
ignore_sessions=True,
normalize='pct'), Fex)
# Test ValueError
with pytest.raises(ValueError):
dat.baseline(baseline='BadValue')
# Test summary
dat2 = dat.loc[:, ['Positive', 'Negative']].interpolate()
out = dat2.extract_summary(min=True, max=True, mean=True)
assert len(out) == len(np.unique(dat2.sessions))
assert np.array_equal(out.sessions, np.unique(dat2.sessions))
assert out.sampling_freq == dat2.sampling_freq
assert dat2.shape[1] * 3 == out.shape[1]
out = dat2.extract_summary(min=True,
max=True,
mean=True,
ignore_sessions=True)
assert len(out) == 1
assert dat2.shape[1] * 3 == out.shape[1]
# Check if file is missing columns
data_bad = dat.iloc[:, 0:10]
with pytest.raises(Exception):
_check_if_fex(data_bad, imotions_columns)
# Check if file has too many columns
data_bad = dat.copy()
data_bad['Test'] = 0
with pytest.raises(Exception):
_check_if_fex(data_bad, imotions_columns)
# Test clean
assert isinstance(dat.clean(), Fex)
assert dat.clean().columns is dat.columns
assert dat.clean().sampling_freq == dat.sampling_freq
# Test Decompose
n_components = 3
stats = dat.decompose(algorithm='pca', axis=1, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='ica', axis=1, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
new_dat = dat + 100
stats = new_dat.decompose(algorithm='nnmf',
axis=1,
n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='fa', axis=1, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='pca', axis=0, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='ica', axis=0, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
new_dat = dat + 100
stats = new_dat.decompose(algorithm='nnmf',
axis=0,
n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='fa', axis=0, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
def test_fex():
# For iMotions-FACET data files
# test reading iMotions file < version 6
dat = Fex(read_facet(join(get_test_data_path(), 'iMotions_Test_v2.txt')),
sampling_freq=30)
# test reading iMotions file > version 6
filename = join(get_test_data_path(), 'iMotions_Test.txt')
df = read_facet(filename)
sessions = np.array([[x] * 10
for x in range(1 + int(len(df) / 10))]).flatten()[:-1]
dat = Fex(df, sampling_freq=30, sessions=sessions)
# Test Session ValueError
with pytest.raises(ValueError):
Fex(df, sampling_freq=30, sessions=sessions[:10])
# Test KeyError
with pytest.raises(KeyError):
Fex(read_facet(filename, features=['NotHere']), sampling_freq=30)
# Test length
assert len(dat) == 519
# Test Info
assert isinstance(dat.info(), str)
# Test sessions generator
assert len(np.unique(dat.sessions)) == len([x for x in dat.itersessions()])
# Test metadata propagation
assert dat['Joy'].sampling_freq == dat.sampling_freq
assert dat.iloc[:, 0].sampling_freq == dat.sampling_freq
# Test Downsample
assert len(dat.downsample(target=10)) == 52
# Test upsample
assert len(dat.upsample(target=60, target_type='hz')) == (len(dat) - 1) * 2
# Test interpolation
assert np.sum(dat.interpolate(method='linear').isnull().sum() == 0) == len(
dat.columns)
dat = dat.interpolate(method='linear')
# Test distance
d = dat.distance()
assert isinstance(d, Adjacency)
assert d.square_shape()[0] == len(dat)
# Test Copy
assert isinstance(dat.copy(), Fex)
assert dat.copy().sampling_freq == dat.sampling_freq
# Test baseline
assert isinstance(dat.baseline(baseline='median'), Fex)
assert isinstance(dat.baseline(baseline='mean'), Fex)
assert isinstance(dat.baseline(baseline='begin'), Fex)
assert isinstance(dat.baseline(baseline=dat.mean()), Fex)
assert isinstance(dat.baseline(baseline='median', ignore_sessions=True),
Fex)
assert isinstance(dat.baseline(baseline='mean', ignore_sessions=True), Fex)
assert isinstance(dat.baseline(baseline=dat.mean(), ignore_sessions=True),
Fex)
assert isinstance(dat.baseline(baseline='median', normalize='pct'), Fex)
assert isinstance(dat.baseline(baseline='mean', normalize='pct'), Fex)
assert isinstance(dat.baseline(baseline=dat.mean(), normalize='pct'), Fex)
assert isinstance(
dat.baseline(baseline='median', ignore_sessions=True, normalize='pct'),
Fex)
assert isinstance(
dat.baseline(baseline='mean', ignore_sessions=True, normalize='pct'),
Fex)
assert isinstance(
dat.baseline(baseline=dat.mean(),
ignore_sessions=True,
normalize='pct'), Fex)
# Test ValueError
with pytest.raises(ValueError):
dat.baseline(baseline='BadValue')
# Test summary
dat2 = dat.loc[:, ['Positive', 'Negative']].interpolate()
out = dat2.extract_summary(min=True, max=True, mean=True)
assert len(out) == len(np.unique(dat2.sessions))
assert np.array_equal(out.sessions, np.unique(dat2.sessions))
assert out.sampling_freq == dat2.sampling_freq
assert dat2.shape[1] * 3 == out.shape[1]
out = dat2.extract_summary(min=True,
max=True,
mean=True,
ignore_sessions=True)
assert len(out) == 1
assert dat2.shape[1] * 3 == out.shape[1]
# Check if file is missing columns
data_bad = dat.iloc[:, 0:10]
with pytest.raises(Exception):
_check_if_fex(data_bad, imotions_columns)
# Check if file has too many columns
data_bad = dat.copy()
data_bad['Test'] = 0
with pytest.raises(Exception):
_check_if_fex(data_bad, imotions_columns)
# Test clean
assert isinstance(dat.clean(), Fex)
assert dat.clean().columns is dat.columns
assert dat.clean().sampling_freq == dat.sampling_freq
# Test Decompose
n_components = 3
stats = dat.decompose(algorithm='pca', axis=1, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='ica', axis=1, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
new_dat = dat + 100
stats = new_dat.decompose(algorithm='nnmf',
axis=1,
n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='fa', axis=1, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='pca', axis=0, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='ica', axis=0, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
new_dat = dat + 100
stats = new_dat.decompose(algorithm='nnmf',
axis=0,
n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='fa', axis=0, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
def test_plot_face():
# test plotting method
fx = Fex(filename=join(get_test_data_path(), 'iMotions_Test_v6.txt'),
sampling_freq=30,
detector='FACET')
fx = fx.read_file()
ax = fx.plot_aus(row_n=0)
assert_plot_shape(ax)
plt.close()
fx = Fex(filename=join(get_test_data_path(), 'OpenFace_Test.csv'),
sampling_freq=30,
detector='OpenFace')
fx = fx.read_file()
ax = fx.plot_aus(row_n=0)
assert_plot_shape(ax)
plt.close()
fx = Fex(filename=join(get_test_data_path(),
'sample_affectiva-api-app_output.json'),
sampling_freq=30,
detector='Affectiva')
fx = fx.read_file(orig_cols=False)
ax = fx.plot_aus(row_n=0)
assert_plot_shape(ax)
plt.close()
# test plot in util
plot_face()
assert_plot_shape(plt.gca())
plt.close()
plot_face(au=au, vectorfield={'reference': predict(au2)})
assert_plot_shape(plt.gca())
plt.close()
with pytest.raises(ValueError):
plot_face(model=au, au=au, vectorfield={'reference': predict(au2)})
with pytest.raises(ValueError):
plot_face(model=au, au=au, vectorfield=[])
with pytest.raises(ValueError):
plot_face(model=au, au=au, vectorfield={'noreference': predict(au2)})
def test_fextractor():
filename = os.path.join(get_test_data_path(), "iMotions_Test_v6.txt")
df = read_facet(filename)
sessions = np.array([[x] * 10
for x in range(1 + int(len(df) / 10))]).flatten()[:-1]
dat = Fex(df, sampling_freq=30, sessions=sessions)
dat = dat[[
"Joy",
"Anger",
"Surprise",
"Fear",
"Contempt",
"Disgust",
"Sadness",
"Confusion",
"Frustration",
"Neutral",
"Positive",
"Negative",
]]
# Test Fextractor class
extractor = Fextractor()
dat = dat.interpolate() # interpolate data to get rid of NAs
f = 0.5
num_cyc = 3 # for wavelet extraction
# Test each extraction method
extractor.mean(fex_object=dat)
extractor.max(fex_object=dat)
extractor.min(fex_object=dat)
# boft needs a groupby function.
extractor.multi_wavelet(fex_object=dat)
extractor.wavelet(fex_object=dat, freq=f, num_cyc=num_cyc)
# Test ValueError
with pytest.raises(ValueError):
extractor.wavelet(fex_object=dat,
freq=f,
num_cyc=num_cyc,
mode="BadValue")
# Test Fextracor merge method
newdat = extractor.merge(out_format="long")
assert newdat["sessions"].nunique() == 52
assert isinstance(newdat, DataFrame)
assert len(extractor.merge(out_format="long")) == 7488
assert len(extractor.merge(out_format="wide")) == 52
# Test summary method
extractor = Fextractor()
dat2 = dat.loc[:, ["Positive", "Negative"]].interpolate()
extractor.summary(fex_object=dat2, min=True, max=True, mean=True)
# [Pos, Neg] * [mean, max, min] + ['sessions']
assert extractor.merge(out_format="wide").shape[1] == dat2.shape[1] * 3 + 1
# Test wavelet extraction
extractor = Fextractor()
extractor.wavelet(fex_object=dat,
freq=f,
num_cyc=num_cyc,
ignore_sessions=False)
extractor.wavelet(fex_object=dat,
freq=f,
num_cyc=num_cyc,
ignore_sessions=True)
wavelet = extractor.extracted_features[0] # ignore_sessions = False
assert wavelet.sampling_freq == dat.sampling_freq
assert len(wavelet) == len(dat)
wavelet = extractor.extracted_features[1] # ignore_sessions = True
assert wavelet.sampling_freq == dat.sampling_freq
assert len(wavelet) == len(dat)
assert np.array_equal(wavelet.sessions, dat.sessions)
for i in ["filtered", "phase", "magnitude", "power"]:
extractor = Fextractor()
extractor.wavelet(fex_object=dat,
freq=f,
num_cyc=num_cyc,
ignore_sessions=True,
mode=i)
wavelet = extractor.extracted_features[0]
assert wavelet.sampling_freq == dat.sampling_freq
assert len(wavelet) == len(dat)
# Test multi wavelet
dat2 = dat.loc[:, ["Positive", "Negative"]].interpolate()
n_bank = 4
extractor = Fextractor()
extractor.multi_wavelet(
fex_object=dat2,
min_freq=0.1,
max_freq=2,
bank=n_bank,
mode="power",
ignore_sessions=False,
)
out = extractor.extracted_features[0]
assert n_bank * dat2.shape[1] == out.shape[1]
assert len(out) == len(dat2)
assert np.array_equal(out.sessions, dat2.sessions)
assert out.sampling_freq == dat2.sampling_freq
# Test Bag Of Temporal Features Extraction
filename = os.path.join(get_test_data_path(), "iMotions_Test_v6.txt")
facet = Fex(filename=filename, sampling_freq=30, detector="FACET")
facet = facet.read_file()
facet_filled = facet.fillna(0)
facet_filled = facet_filled[[
"Joy",
"Anger",
"Surprise",
"Fear",
"Contempt",
"Disgust",
"Sadness",
"Confusion",
"Frustration",
"Neutral",
"Positive",
"Negative",
]]
# assert isinstance(facet_filled,Facet)
extractor = Fextractor()
extractor.boft(facet_filled)
assert isinstance(extractor.extracted_features[0], DataFrame)
filters, histograms = 8, 12
assert (extractor.extracted_features[0].shape[1] ==
facet_filled.columns.shape[0] * filters * histograms)
def test_fex():
with pytest.raises(Exception):
fex = Fex().read_feat()
with pytest.raises(Exception):
fex = Fex().read_facet()
with pytest.raises(Exception):
fex = Fex().read_openface()
with pytest.raises(Exception):
fex = Fex().read_affectiva()
# For iMotions-FACET data files
# test reading iMotions file < version 6
filename = os.path.join(get_test_data_path(), "iMotions_Test_v5.txt")
dat = Fex(read_facet(filename), sampling_freq=30)
# test reading iMotions file > version 6
filename = os.path.join(get_test_data_path(), "iMotions_Test_v6.txt")
df = read_facet(filename)
# Test slicing functions.
assert df.aus().shape == (519, 20)
assert df.emotions().shape == (519, 12)
assert df.facebox().shape == (519, 4)
assert df.time().shape[-1] == 4
assert df.design().shape[-1] == 4
# Test metadata propagation to sliced series
assert df.iloc[0].aus().shape == (20, )
assert df.iloc[0].emotions().shape == (12, )
assert df.iloc[0].facebox().shape == (4, )
assert df.iloc[0].time().shape == (4, )
assert df.iloc[0].design().shape == (4, )
sessions = np.array([[x] * 10
for x in range(1 + int(len(df) / 10))]).flatten()[:-1]
dat = Fex(df, sampling_freq=30, sessions=sessions)
dat = dat[[
"Joy",
"Anger",
"Surprise",
"Fear",
"Contempt",
"Disgust",
"Sadness",
"Confusion",
"Frustration",
"Neutral",
"Positive",
"Negative",
]]
# Test Session ValueError
with pytest.raises(ValueError):
Fex(df, sampling_freq=30, sessions=sessions[:10])
# Test KeyError
with pytest.raises(KeyError):
Fex(read_facet(filename, features=["NotHere"]), sampling_freq=30)
# Test length
assert len(dat) == 519
# Test sessions generator
assert len(np.unique(dat.sessions)) == len([x for x in dat.itersessions()])
# Test metadata propagation
assert dat[["Joy"]].sampling_freq == dat.sampling_freq
assert dat.iloc[:, 0].sampling_freq == dat.sampling_freq
assert dat.iloc[0, :].sampling_freq == dat.sampling_freq
assert dat.loc[[0], :].sampling_freq == dat.sampling_freq
assert dat.loc[:, ["Joy"]].sampling_freq == dat.sampling_freq
# assert dat.loc[0].sampling_freq == dat.sampling_freq # DOES NOT WORK YET
# Test Downsample
assert len(dat.downsample(target=10)) == 52
# Test upsample
assert len(dat.upsample(target=60, target_type="hz")) == (len(dat) - 1) * 2
# Test interpolation
assert (dat.interpolate(method="linear").isnull().sum()["Positive"] <
dat.isnull().sum()["Positive"])
dat = dat.interpolate(method="linear")
# Test distance
d = dat[["Positive"]].distance()
assert isinstance(d, Adjacency)
assert d.square_shape()[0] == len(dat)
# Test Copy
assert isinstance(dat.copy(), Fex)
assert dat.copy().sampling_freq == dat.sampling_freq
# Test rectification
rectified = df.rectification()
assert (df[df.au_columns].isna().sum()[0] <
rectified[rectified.au_columns].isna().sum()[0])
# Test pspi
assert len(df.calc_pspi()) == len(df)
# Test baseline
assert isinstance(dat.baseline(baseline="median"), Fex)
assert isinstance(dat.baseline(baseline="mean"), Fex)
assert isinstance(dat.baseline(baseline="begin"), Fex)
assert isinstance(dat.baseline(baseline=dat.mean()), Fex)
assert isinstance(dat.baseline(baseline="median", ignore_sessions=True),
Fex)
assert isinstance(dat.baseline(baseline="mean", ignore_sessions=True), Fex)
assert isinstance(dat.baseline(baseline=dat.mean(), ignore_sessions=True),
Fex)
assert isinstance(dat.baseline(baseline="median", normalize="pct"), Fex)
assert isinstance(dat.baseline(baseline="mean", normalize="pct"), Fex)
assert isinstance(dat.baseline(baseline=dat.mean(), normalize="pct"), Fex)
assert isinstance(
dat.baseline(baseline="median", ignore_sessions=True, normalize="pct"),
Fex)
assert isinstance(
dat.baseline(baseline="mean", ignore_sessions=True, normalize="pct"),
Fex)
assert isinstance(
dat.baseline(baseline=dat.mean(),
ignore_sessions=True,
normalize="pct"), Fex)
# Test ValueError
with pytest.raises(ValueError):
dat.baseline(baseline="BadValue")
# Test summary
dat2 = dat.loc[:, ["Positive", "Negative"]].interpolate()
out = dat2.extract_summary(min=True, max=True, mean=True)
assert len(out) == len(np.unique(dat2.sessions))
assert np.array_equal(out.sessions, np.unique(dat2.sessions))
assert out.sampling_freq == dat2.sampling_freq
assert dat2.shape[1] * 3 == out.shape[1]
out = dat2.extract_summary(min=True,
max=True,
mean=True,
ignore_sessions=True)
assert len(out) == 1
assert dat2.shape[1] * 3 == out.shape[1]
# Check if file is missing columns
data_bad = dat.iloc[:, 0:10]
with pytest.raises(Exception):
_check_if_fex(data_bad, imotions_columns)
# Check if file has too many columns
data_bad = dat.copy()
data_bad["Test"] = 0
with pytest.raises(Exception):
_check_if_fex(data_bad, imotions_columns)
# Test clean
assert isinstance(dat.clean(), Fex)
assert dat.clean().columns is dat.columns
assert dat.clean().sampling_freq == dat.sampling_freq
# Test Decompose
n_components = 3
stats = dat.decompose(algorithm="pca", axis=1, n_components=n_components)
assert n_components == stats["components"].shape[1]
assert n_components == stats["weights"].shape[1]
stats = dat.decompose(algorithm="ica", axis=1, n_components=n_components)
assert n_components == stats["components"].shape[1]
assert n_components == stats["weights"].shape[1]
new_dat = dat + 100
stats = new_dat.decompose(algorithm="nnmf",
axis=1,
n_components=n_components)
assert n_components == stats["components"].shape[1]
assert n_components == stats["weights"].shape[1]
stats = dat.decompose(algorithm="fa", axis=1, n_components=n_components)
assert n_components == stats["components"].shape[1]
assert n_components == stats["weights"].shape[1]
stats = dat.decompose(algorithm="pca", axis=0, n_components=n_components)
assert n_components == stats["components"].shape[1]
assert n_components == stats["weights"].shape[1]
stats = dat.decompose(algorithm="ica", axis=0, n_components=n_components)
assert n_components == stats["components"].shape[1]
assert n_components == stats["weights"].shape[1]
new_dat = dat + 100
stats = new_dat.decompose(algorithm="nnmf",
axis=0,
n_components=n_components)
assert n_components == stats["components"].shape[1]
assert n_components == stats["weights"].shape[1]
stats = dat.decompose(algorithm="fa", axis=0, n_components=n_components)
assert n_components == stats["components"].shape[1]
assert n_components == stats["weights"].shape[1]
def test_fextractor():
filename = join(get_test_data_path(), 'iMotions_Test.txt')
df = read_facet(filename)
sessions = np.array([[x] * 10
for x in range(1 + int(len(df) / 10))]).flatten()[:-1]
dat = Fex(df, sampling_freq=30, sessions=sessions)
# Test Fextractor class
extractor = Fextractor()
dat = dat.interpolate() # interpolate data to get rid of NAs
f = .5
num_cyc = 3 # for wavelet extraction
# Test each extraction method
extractor.mean(fex_object=dat)
extractor.max(fex_object=dat)
extractor.min(fex_object=dat)
#extractor.boft(fex_object=dat, min_freq=.01, max_freq=.20, bank=1)
extractor.multi_wavelet(fex_object=dat)
extractor.wavelet(fex_object=dat, freq=f, num_cyc=num_cyc)
# Test ValueError
with pytest.raises(ValueError):
extractor.wavelet(fex_object=dat,
freq=f,
num_cyc=num_cyc,
mode='BadValue')
# Test Fextracor merge method
newdat = extractor.merge(out_format='long')
assert newdat['sessions'].nunique() == 52
assert isinstance(newdat, DataFrame)
assert len(extractor.merge(out_format='long')) == 24960
assert len(extractor.merge(out_format='wide')) == 52
# Test summary method
extractor = Fextractor()
dat2 = dat.loc[:, ['Positive', 'Negative']].interpolate()
extractor.summary(fex_object=dat2, min=True, max=True, mean=True)
# [Pos, Neg] * [mean, max, min] + ['sessions']
assert extractor.merge(out_format='wide').shape[1] == dat2.shape[1] * 3 + 1
# Test wavelet extraction
extractor = Fextractor()
extractor.wavelet(fex_object=dat,
freq=f,
num_cyc=num_cyc,
ignore_sessions=False)
extractor.wavelet(fex_object=dat,
freq=f,
num_cyc=num_cyc,
ignore_sessions=True)
wavelet = extractor.extracted_features[0] # ignore_sessions = False
assert wavelet.sampling_freq == dat.sampling_freq
assert len(wavelet) == len(dat)
wavelet = extractor.extracted_features[1] # ignore_sessions = True
assert wavelet.sampling_freq == dat.sampling_freq
assert len(wavelet) == len(dat)
assert np.array_equal(wavelet.sessions, dat.sessions)
for i in ['filtered', 'phase', 'magnitude', 'power']:
extractor = Fextractor()
extractor.wavelet(fex_object=dat,
freq=f,
num_cyc=num_cyc,
ignore_sessions=True,
mode=i)
wavelet = extractor.extracted_features[0]
assert wavelet.sampling_freq == dat.sampling_freq
assert len(wavelet) == len(dat)
# Test multi wavelet
dat2 = dat.loc[:, ['Positive', 'Negative']].interpolate()
n_bank = 4
extractor = Fextractor()
extractor.multi_wavelet(fex_object=dat2,
min_freq=.1,
max_freq=2,
bank=n_bank,
mode='power',
ignore_sessions=False)
out = extractor.extracted_features[0]
assert n_bank * dat2.shape[1] == out.shape[1]
assert len(out) == len(dat2)
assert np.array_equal(out.sessions, dat2.sessions)
assert out.sampling_freq == dat2.sampling_freq
# Test Bag Of Temporal Features Extraction
filename = join(get_test_data_path(), 'iMotions_Test.txt')
facet = Facet(filename=filename, sampling_freq=30)
facet.read_file()
facet_filled = facet.fillna(0)
assert isinstance(facet_filled, Facet)
extractor = Fextractor()
extractor.boft(facet_filled)
assert isinstance(extractor.extracted_features[0], DataFrame)
filters, histograms = 8, 12
assert extractor.extracted_features[0].shape[
1] == facet.columns.shape[0] * filters * histograms
def test_info(capsys):
importantstring = "ThisStringMustBeIncluded"
fex = Fex(filename=importantstring)
fex.info()
captured = capsys.readouterr()
assert importantstring in captured.out
def test_fex(tmpdir):
# For iMotions-FACET data files
# test reading iMotions file < version 6
dat = Fex(read_facet(join(get_test_data_path(), 'iMotions_Test_v2.txt')),
sampling_freq=30)
# test reading iMotions file > version 6
filename = join(get_test_data_path(), 'iMotions_Test.txt')
df = read_facet(filename)
sessions = np.array([[x] * 10
for x in range(1 + int(len(df) / 10))]).flatten()[:-1]
dat = Fex(df, sampling_freq=30, sessions=sessions)
# Test KeyError
class MyTestCase(unittest.TestCase):
def test1(self):
with self.assertRaises(KeyError):
Fex(read_facet(filename, features=['NotHere']),
sampling_freq=30)
# Test length
assert len(dat) == 519
# Test Info
assert isinstance(dat.info(), str)
# Test sessions generator
assert len(np.unique(dat.sessions)) == len([x for x in dat.itersessions()])
# Test metadata propagation
assert dat['Joy'].sampling_freq == dat.sampling_freq
assert dat.iloc[:, 0].sampling_freq == dat.sampling_freq
# Test Downsample
assert len(dat.downsample(target=10)) == 52
# Test upsample
assert len(dat.upsample(target=60, target_type='hz')) == (len(dat) - 1) * 2
# Test interpolation
assert np.sum(dat.interpolate(method='linear').isnull().sum() == 0) == len(
dat.columns)
dat = dat.interpolate(method='linear')
# Test distance
d = dat.distance()
assert isinstance(d, Adjacency)
assert d.square_shape()[0] == len(dat)
# Test Copy
assert isinstance(dat.copy(), Fex)
assert dat.copy().sampling_freq == dat.sampling_freq
# Test baseline
assert isinstance(dat.baseline(baseline='median'), Fex)
assert isinstance(dat.baseline(baseline='mean'), Fex)
assert isinstance(dat.baseline(baseline=dat.mean()), Fex)
assert isinstance(dat.baseline(baseline='median', ignore_sessions=True),
Fex)
assert isinstance(dat.baseline(baseline='mean', ignore_sessions=True), Fex)
assert isinstance(dat.baseline(baseline=dat.mean(), ignore_sessions=True),
Fex)
assert isinstance(dat.baseline(baseline='median', normalize='pct'), Fex)
assert isinstance(dat.baseline(baseline='mean', normalize='pct'), Fex)
assert isinstance(dat.baseline(baseline=dat.mean(), normalize='pct'), Fex)
assert isinstance(
dat.baseline(baseline='median', ignore_sessions=True, normalize='pct'),
Fex)
assert isinstance(
dat.baseline(baseline='mean', ignore_sessions=True, normalize='pct'),
Fex)
assert isinstance(
dat.baseline(baseline=dat.mean(),
ignore_sessions=True,
normalize='pct'), Fex)
# Test facet subclass
facet = Facet(filename=filename, sampling_freq=30)
facet.read_file()
assert len(facet) == 519
# Test PSPI calculation
assert len(facet.calc_pspi()) == len(facet)
# Test Fextractor class
extractor = Fextractor()
dat = dat.interpolate() # interpolate data to get rid of NAs
f = .5
num_cyc = 3 # for wavelet extraction
# Test each extraction method
extractor.mean(fex_object=dat)
extractor.max(fex_object=dat)
extractor.min(fex_object=dat)
#extractor.boft(fex_object=dat, min_freq=.01, max_freq=.20, bank=1)
extractor.multi_wavelet(fex_object=dat)
extractor.wavelet(fex_object=dat, freq=f, num_cyc=num_cyc)
# Test Fextracor merge method
newdat = extractor.merge(out_format='long')
assert newdat['sessions'].nunique() == 52
assert isinstance(newdat, DataFrame)
assert len(extractor.merge(out_format='long')) == 24960
assert len(extractor.merge(out_format='wide')) == 52
# Test wavelet extraction
extractor = Fextractor()
extractor.wavelet(fex_object=dat,
freq=f,
num_cyc=num_cyc,
ignore_sessions=False)
extractor.wavelet(fex_object=dat,
freq=f,
num_cyc=num_cyc,
ignore_sessions=True)
wavelet = extractor.extracted_features[0] # ignore_sessions = False
assert wavelet.sampling_freq == dat.sampling_freq
assert len(wavelet) == len(dat)
wavelet = extractor.extracted_features[1] # ignore_sessions = True
assert wavelet.sampling_freq == dat.sampling_freq
assert len(wavelet) == len(dat)
assert np.array_equal(wavelet.sessions, dat.sessions)
for i in ['filtered', 'phase', 'magnitude', 'power']:
extractor = Fextractor()
extractor.wavelet(fex_object=dat,
freq=f,
num_cyc=num_cyc,
ignore_sessions=True,
mode=i)
wavelet = extractor.extracted_features[0]
assert wavelet.sampling_freq == dat.sampling_freq
assert len(wavelet) == len(dat)
# Test multi wavelet
dat2 = dat.loc[:, ['Positive', 'Negative']].interpolate()
n_bank = 4
extractor = Fextractor()
extractor.multi_wavelet(fex_object=dat2,
min_freq=.1,
max_freq=2,
bank=n_bank,
mode='power',
ignore_sessions=False)
out = extractor.extracted_features[0]
assert n_bank * dat2.shape[1] == out.shape[1]
assert len(out) == len(dat2)
assert np.array_equal(out.sessions, dat2.sessions)
assert out.sampling_freq == dat2.sampling_freq
# Test Bag Of Temporal Features Extraction
facet_filled = facet.fillna(0)
assert isinstance(facet_filled, Facet)
extractor = Fextractor()
extractor.boft(facet_filled)
assert isinstance(extractor.extracted_features[0], DataFrame)
filters, histograms = 8, 12
assert extractor.extracted_features[0].shape[
1] == facet.columns.shape[0] * filters * histograms
# Test mean, min, and max Features Extraction
# assert isinstance(facet_filled.extract_mean(), Facet)
# assert isinstance(facet_filled.extract_min(), Facet)
# assert isinstance(facet_filled.extract_max(), Facet)
# Test if a method returns subclass.
facet = facet.downsample(target=10, target_type='hz')
assert isinstance(facet, Facet)
### Test Openface importer and subclass ###
# For OpenFace data file
filename = join(get_test_data_path(), 'OpenFace_Test.csv')
openface = Fex(read_openface(filename), sampling_freq=30)
# Test KeyError
class MyTestCase(unittest.TestCase):
def test1(self):
with self.assertRaises(KeyError):
Fex(read_openface(filename, features=['NotHere']),
sampling_freq=30)
# Test length
assert len(openface) == 100
# Test loading from filename
openface = Openface(filename=filename, sampling_freq=30)
openface.read_file()
# Test length?
assert len(openface) == 100
# Test PSPI calculation b/c diff from facet
assert len(openface.calc_pspi()) == len(openface)
# Test if a method returns subclass.
openface = openface.downsample(target=10, target_type='hz')
assert isinstance(openface, Openface)
# Check if file is missing columns
data_bad = dat.iloc[:, 0:10]
with pytest.raises(Exception):
_check_if_fex(data_bad, imotions_columns)
# Check if file has too many columns
data_bad = dat.copy()
data_bad['Test'] = 0
with pytest.raises(Exception):
_check_if_fex(data_bad, imotions_columns)
# Test clean
assert isinstance(dat.clean(), Fex)
assert dat.clean().columns is dat.columns
assert dat.clean().sampling_freq == dat.sampling_freq
# Test Decompose
n_components = 3
stats = dat.decompose(algorithm='pca', axis=1, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='ica', axis=1, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
new_dat = dat + 100
stats = new_dat.decompose(algorithm='nnmf',
axis=1,
n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='fa', axis=1, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='pca', axis=0, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='ica', axis=0, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
new_dat = dat + 100
stats = new_dat.decompose(algorithm='nnmf',
axis=0,
n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
stats = dat.decompose(algorithm='fa', axis=0, n_components=n_components)
assert n_components == stats['components'].shape[1]
assert n_components == stats['weights'].shape[1]
def test1(self):
with self.assertRaises(KeyError):
Fex(read_facet(filename, features=['NotHere']),
sampling_freq=30)
def test_feat():
filename = os.path.join(get_test_data_path(), 'output.csv')
fex = Fex(filename=filename, detector='Feat')
fex.read_file()
def test_affectiva():
filename = join(get_test_data_path(),
'sample_affectiva-api-app_output.json')
affdex = Fex(read_affectiva(filename), sampling_freq=1)
assert affdex.shape[1] == 32
def detect_video(self,
inputFname,
outputFname=None,
skip_frames=1,
verbose=False):
"""Detects FEX from a video file.
Args:
inputFname (str): Path to video file
outputFname (str, optional): Path to output file. Defaults to None.
skip_frames (int, optional): Number of every other frames to skip for speed or if not all frames need to be processed. Defaults to 1.
Returns:
dataframe: Prediction results dataframe if outputFname is None. Returns True if outputFname is specified.
"""
self.info["inputFname"] = inputFname
self.info["outputFname"] = outputFname
init_df = pd.DataFrame(columns=self["output_columns"])
if outputFname:
init_df.to_csv(outputFname, index=False, header=True)
cap = cv2.VideoCapture(inputFname)
length = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
frames_to_process = int(np.ceil(length / skip_frames))
# Determine whether to use multiprocessing.
n_jobs = self["n_jobs"]
if n_jobs == -1:
thread_num = cv2.getNumberOfCPUs() # get available cpus
else:
thread_num = n_jobs
if verbose:
print(f"Using {thread_num} cpus")
pool = ThreadPool(processes=thread_num)
pending_task = deque()
counter = 0
processed_frames = 0
frame_got = True
detected_faces = []
if verbose:
print("Processing video.")
# single core
while True:
frame_got, frame = cap.read()
if counter % skip_frames == 0:
df = self.process_frame(frame, counter=counter)
df["input"] = inputFname
if outputFname:
df[init_df.columns].to_csv(outputFname,
index=False,
header=False,
mode="a")
else:
init_df = pd.concat([init_df, df[init_df.columns]], axis=0)
counter = counter + 1
if not frame_got:
break
cap.release()
if outputFname:
return True
else:
return Fex(
init_df,
filename=inputFname,
au_columns=self["au_presence_columns"],
emotion_columns=FEAT_EMOTION_COLUMNS,
facebox_columns=FEAT_FACEBOX_COLUMNS,
landmark_columns=openface_2d_landmark_columns,
time_columns=FEAT_TIME_COLUMNS,
detector="Feat",
)
def detect_image(self,
inputFname,
batch_size=5,
outputFname=None,
verbose=False,
singleframe4error=False):
"""Detects FEX from an image file.
Args:
inputFname (list of str): Path to a list of paths to image files.
bacth_size (int, optional): how many batches of images you want to run at one shot. Larger gives faster speed but is more memory-consuming
outputFname (str, optional): Path to output file. Defaults to None.
singleframe4error (bool, default = False): When set True, when exception occurs inside a batch, instead of nullify the whole batch, process each img in
batch individually
Rseturns:
Fex: Prediction results dataframe if outputFname is None. Returns True if outputFname is specified.
"""
assert (
type(inputFname) == str or type(inputFname) == list
), "inputFname must be a string path to image or list of image paths"
if type(inputFname) == str:
inputFname = [inputFname]
for inputF in inputFname:
if not os.path.exists(inputF):
raise FileNotFoundError(f"File {inputF} not found.")
self.info["inputFname"] = inputFname
init_df = pd.DataFrame(columns=self["output_columns"])
if outputFname:
init_df.to_csv(outputFname, index=False, header=True)
counter = 0
concat_frame = None
input_names = []
while counter < len(inputFname):
#if counter % skip_frames == 0:
frame = np.expand_dims(cv2.imread(inputFname[counter]), 0)
if concat_frame is None:
concat_frame = frame
tmp_counter = counter
else:
concat_frame = np.concatenate([concat_frame, frame], 0)
input_names.append(inputFname[counter])
counter = counter + 1
if (counter % batch_size == 0) and (concat_frame is not None):
if singleframe4error:
try:
df, _ = self.process_frame(
concat_frame,
counter=tmp_counter,
singleframe4error=singleframe4error)
except FaceDetectionError:
df = None
for id_fr in range(concat_frame.shape[0]):
tmp_df, _ = self.process_frame(
concat_frame[id_fr:(id_fr + 1)],
counter=tmp_counter,
singleframe4error=False)
tmp_counter += 1
if df is None:
df = tmp_df
else:
df = pd.concat((df, tmp_df), 0)
else:
df, _ = self.process_frame(concat_frame,
counter=tmp_counter)
df["input"] = input_names
if outputFname:
df[init_df.columns].to_csv(outputFname,
index=False,
header=False,
mode="a")
else:
init_df = pd.concat([init_df, df[init_df.columns]], axis=0)
concat_frame = None
tmp_counter = None
input_names = []
if len(inputFname) % batch_size != 0:
# process remaining frames
if concat_frame is not None:
if singleframe4error:
try:
df, _ = self.process_frame(concat_frame,
counter=tmp_counter)
except FaceDetectionError:
df = None
for id_fr in range(concat_frame.shape[0]):
tmp_df, _ = self.process_frame(
concat_frame[id_fr:(id_fr + 1)],
counter=tmp_counter,
singleframe4error=False)
tmp_counter += 1
if df is None:
df = tmp_df
else:
df = pd.concat((df, tmp_df), 0)
else:
df, _ = self.process_frame(concat_frame,
counter=tmp_counter)
df["input"] = input_names
if outputFname:
df[init_df.columns].to_csv(outputFname,
index=False,
header=False,
mode="a")
else:
init_df = pd.concat([init_df, df[init_df.columns]], axis=0)
if outputFname:
return True
else:
return Fex(
init_df,
filename=inputFname,
au_columns=self['au_presence_columns'],
emotion_columns=FEAT_EMOTION_COLUMNS,
facebox_columns=FEAT_FACEBOX_COLUMNS,
landmark_columns=openface_2d_landmark_columns,
facepose_columns=FACET_FACEPOSE_COLUMNS,
time_columns=FACET_TIME_COLUMNS,
detector="Feat",
)
def detect_video(self,
inputFname,
batch_size=5,
outputFname=None,
skip_frames=1,
verbose=False,
singleframe4error=False):
"""Detects FEX from a video file.
Args:
inputFname (str): Path to video file
outputFname (str, optional): Path to output file. Defaults to None.
bacth_size (int, optional): how many batches of images you want to run at one shot. Larger gives faster speed but is more memory-consuming
skip_frames (int, optional): Number of every other frames to skip for speed or if not all frames need to be processed. Defaults to 1.
singleframe4error (bool, default = False): When set True, when exception occurs inside a batch, instead of nullify the whole batch, process each img in
batch individually
Returns:
dataframe: Prediction results dataframe if outputFname is None. Returns True if outputFname is specified.
"""
self.info["inputFname"] = inputFname
self.info["outputFname"] = outputFname
init_df = pd.DataFrame(columns=self["output_columns"])
if outputFname:
init_df.to_csv(outputFname, index=False, header=True)
cap = cv2.VideoCapture(inputFname)
length = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
frames_to_process = int(np.ceil(length / skip_frames))
counter = 0
frame_got = True
if verbose:
print("Processing video.")
# single core
concat_frame = None
while True:
frame_got, frame = cap.read()
if frame_got:
if counter % skip_frames == 0:
# if the
if concat_frame is None:
concat_frame = np.expand_dims(frame, 0)
tmp_counter = counter
else:
concat_frame = np.concatenate(
[concat_frame,
np.expand_dims(frame, 0)], 0)
if (concat_frame is not None) and (counter != 0) and (
concat_frame.shape[0] % batch_size
== 0): # I think it's probably this error
if singleframe4error:
try:
df, _ = self.process_frame(
concat_frame,
counter=tmp_counter,
singleframe4error=singleframe4error,
skip_frame_rate=skip_frames)
except FaceDetectionError:
df = None
for id_fr in range(concat_frame.shape[0]):
tmp_df, _ = self.process_frame(
concat_frame[id_fr:(id_fr + 1)],
counter=tmp_counter,
singleframe4error=False,
skip_frame_rate=skip_frames)
tmp_counter += 1
if df is None:
df = tmp_df
else:
df = pd.concat((df, tmp_df), 0)
else:
df, _ = self.process_frame(concat_frame,
counter=tmp_counter,
skip_frame_rate=skip_frames)
df["input"] = inputFname
if outputFname:
df[init_df.columns].to_csv(outputFname,
index=False,
header=False,
mode="a")
else:
init_df = pd.concat([init_df, df[init_df.columns]],
axis=0)
concat_frame = None
tmp_counter = None
counter = counter + 1
else:
# process remaining frames
if concat_frame is not None:
if singleframe4error:
try:
df, _ = self.process_frame(
concat_frame,
counter=tmp_counter,
skip_frame_rate=skip_frames)
except FaceDetectionError:
df = None
for id_fr in range(concat_frame.shape[0]):
tmp_df, _ = self.process_frame(
concat_frame[id_fr:(id_fr + 1)],
counter=tmp_counter,
singleframe4error=False,
skip_frame_rate=skip_frames)
tmp_counter += 1
if df is None:
df = tmp_df
else:
df = pd.concat((df, tmp_df), 0)
else:
df, _ = self.process_frame(concat_frame,
counter=tmp_counter,
skip_frame_rate=skip_frames)
df["input"] = inputFname
if outputFname:
df[init_df.columns].to_csv(outputFname,
index=False,
header=False,
mode="a")
else:
init_df = pd.concat([init_df, df[init_df.columns]],
axis=0)
break
cap.release()
if outputFname:
return True
else:
return Fex(
init_df,
filename=inputFname,
au_columns=self["au_presence_columns"],
emotion_columns=FEAT_EMOTION_COLUMNS,
facebox_columns=FEAT_FACEBOX_COLUMNS,
landmark_columns=openface_2d_landmark_columns,
facepose_columns=FACET_FACEPOSE_COLUMNS,
time_columns=FEAT_TIME_COLUMNS,
detector="Feat",
)
