def donttest_detector_train(
self): # TODO: Cascade training fails for Version OpenCV 2.0
positives = []
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA, "coords.txt"))
n = len(self.eyes.files())
for filename in self.eyes.files()[:n / 2]:
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
faces = self.eyes.getFaces(img.filename)
positives.append(
[os.path.join(SCRAPS_FACE_DATA, img.filename), faces])
neg_files = []
for im_name in os.listdir(NONFACE_DATA):
if im_name[-4:] != ".jpg": continue
neg_files.append(os.path.join(NONFACE_DATA, im_name))
fd = trainHaarClassifier(positives,
neg_files,
nstages=6,
maxtreesplits=0,
max_run_time=300)
fdt = FaceDetectionTest(name='scraps')
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA, "coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth, rects, im=img)
self.assertAlmostEqual(fdt.pos_rate, 0.9942196531791907)
def donttest_detector_train(self): # TODO: Cascade training fails for Version OpenCV 2.0
positives = []
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
n = len(self.eyes.files())
for filename in self.eyes.files()[:n/2]:
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
faces = self.eyes.getFaces(img.filename)
positives.append([os.path.join(SCRAPS_FACE_DATA,img.filename),faces])
neg_files = []
for im_name in os.listdir(NONFACE_DATA):
if im_name[-4:] != ".jpg": continue
neg_files.append(os.path.join(NONFACE_DATA,im_name))
fd = trainHaarClassifier(positives,neg_files,nstages=6,maxtreesplits=0,max_run_time=300)
fdt = FaceDetectionTest(name='scraps')
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth,rects,im=img)
self.assertAlmostEqual( fdt.pos_rate , 0.9942196531791907 )
class _TestSVMEyeDetector(unittest.TestCase):
def setUp(self):
self.images = []
self.names = []
SCRAPS_FACE_DATA = os.path.join(pv.__path__[0], "data",
"csuScrapShots")
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA, "coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
self.images.append(img)
self.names.append(filename)
self.assert_(len(self.images) == 173)
def test_training(self):
'''
This trains the FaceFinder on the scraps database.
'''
#import cProfile
# Load an eyes file
eyes_filename = join(pv.__path__[0], 'data', 'csuScrapShots',
'coords.txt')
#print "Creating eyes File."
eyes_file = EyesFile(eyes_filename)
# Create a face detector
cascade_file = join(pv.__path__[0], 'config',
'facedetector_celebdb2.xml')
#print "Creating a face detector from:",cascade_file
face_detector = CascadeDetector(cascade_file)
image_dir = join(pv.__path__[0], 'data', 'csuScrapShots')
ed = SVMEyeDetectorFromDatabase(eyes_file,
image_dir,
image_ext=".pgm",
face_detector=face_detector,
random_seed=0)
edt = EyeDetectionTest(name='scraps')
#print "Testing..."
for img in self.images:
#print img.filename
faces = ed.detect(img)
#faces = ed.detect(img)
pred_eyes = []
for _, _, pleye, preye in faces:
#detections.append(rect)
pred_eyes.append((pleye, preye))
truth_eyes = self.eyes.getEyes(img.filename)
edt.addSample(truth_eyes, pred_eyes, im=img, annotate=False)
#print edt.createSummary()
self.assertAlmostEqual(edt.face_rate, 0.924855491329, places=3)
def setUp(self):
self.images = []
self.names = []
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA, "coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
self.images.append(img)
self.names.append(filename)
self.assert_(len(self.images) == 173)
class _TestSVMEyeDetector(unittest.TestCase):
def setUp(self):
self.images = []
self.names = []
SCRAPS_FACE_DATA = os.path.join(pv.__path__[0],"data","csuScrapShots")
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
self.images.append(img)
self.names.append(filename)
self.assert_( len(self.images) == 173 )
def test_training(self):
'''
This trains the FaceFinder on the scraps database.
'''
#import cProfile
# Load an eyes file
eyes_filename = join(pv.__path__[0],'data','csuScrapShots','coords.txt')
#print "Creating eyes File."
eyes_file = EyesFile(eyes_filename)
# Create a face detector
cascade_file = join(pv.__path__[0],'config','facedetector_celebdb2.xml')
#print "Creating a face detector from:",cascade_file
face_detector = CascadeDetector(cascade_file)
image_dir = join(pv.__path__[0],'data','csuScrapShots')
ed = SVMEyeDetectorFromDatabase(eyes_file, image_dir, image_ext=".pgm", face_detector=face_detector,random_seed=0)
edt = EyeDetectionTest(name='scraps')
#print "Testing..."
for img in self.images:
#print img.filename
faces = ed.detect(img)
#faces = ed.detect(img)
pred_eyes = []
for _,_,pleye,preye in faces:
#detections.append(rect)
pred_eyes.append((pleye,preye))
truth_eyes = self.eyes.getEyes(img.filename)
edt.addSample(truth_eyes, pred_eyes, im=img, annotate=False)
#print edt.createSummary()
self.assertAlmostEqual( edt.face_rate , 0.924855491329, places = 3 )
def setUp(self):
SCRAPS_FACE_DATA = os.path.join(pv.__path__[0],"data","csuScrapShots")
self.test_images = []
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files()[0:10]:
im = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
eyes = self.eyes.getEyes(filename)
#print eyes
affine = pv.AffineFromPoints(eyes[0][0],eyes[0][1],pv.Point(40,40),pv.Point(88,40),(128,128))
im = affine.transformImage(im)
self.test_images.append(im)
def test_detect_scraps_celeb1(self):
fd = CascadeDetector(CELEB1_CASCADE)
fdt = FaceDetectionTest(name='scraps')
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth,rects,im=img)
self.assertAlmostEqual( fdt.pos_rate , 0.76878612716763006, places = 2 ) # TODO: Version 2 performance is better
def test_face_detection_pickle(self):
fd = CascadeDetector(OPENCV_CASCADE)
fdt = FaceDetectionTest(name='scraps')
data_buffer = pickle.dumps(fd)
fd = pickle.loads(data_buffer)
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth,rects,im=img)
self.assertAlmostEqual( fdt.pos_rate , 0.98265895953757221, places = 2 ) # TODO: Version 2 performance is better
def setUp(self):
self.images = []
self.names = []
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
self.images.append(img)
self.names.append(filename)
self.assert_( len(self.images) == 173 )
def setUp(self):
SCRAPS_FACE_DATA = os.path.join(pv.__path__[0],"data","csuScrapShots")
self.test_images = []
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files()[0:10]:
im = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
eyes = self.eyes.getEyes(filename)
#print eyes
affine = pv.AffineFromPoints(eyes[0][0],eyes[0][1],pv.Point(40,40),pv.Point(88,40),(128,128))
im = affine.transformImage(im)
self.test_images.append(im)
def test_detect_scraps_celeb1(self):
fd = CascadeDetector(CELEB1_CASCADE)
fdt = FaceDetectionTest(name='scraps')
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth,rects,im=img)
self.assertAlmostEqual( fdt.pos_rate , 0.76878612716763006, places = 2 ) # TODO: Version 2 performance is better
def setUp(self):
self.images = []
self.names = []
SCRAPS_FACE_DATA = os.path.join(pv.__path__[0],"data","csuScrapShots")
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
self.images.append(img)
self.names.append(filename)
self.assert_( len(self.images) == 173 )
class _TestFacePCA(unittest.TestCase):
def setUp(self):
self.images = []
self.names = []
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA, "coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
self.images.append(img)
self.names.append(filename)
self.assert_(len(self.images) == 173)
def test_pca_scraps(self):
face_test = FaceRecognitionTest.FaceRecognitionTest(
name='PCA_CSUScraps',
score_type=FaceRecognitionTest.SCORE_TYPE_HIGH)
pca = PCA(drop_front=2, basis_vectors=55)
for im_name in self.eyes.files():
im = pv.Image(os.path.join(SCRAPS_FACE_DATA, im_name + ".pgm"))
rect = self.eyes.getFaces(im_name)
eyes = self.eyes.getEyes(im_name)
pca.addTraining(im, rect=rect[0], eyes=eyes[0])
pca.train()
face_records = {}
for im_name in self.eyes.files():
im = pv.Image(os.path.join(SCRAPS_FACE_DATA, im_name + ".pgm"))
rect = self.eyes.getFaces(im_name)
eyes = self.eyes.getEyes(im_name)
fr = pca.computeFaceRecord(im, rect=rect[0], eyes=eyes[0])
face_records[im_name] = fr
for i_name in face_records.keys():
scores = []
for j_name in face_records.keys():
similarity = pca.similarity(face_records[i_name],
face_records[j_name])
scores.append((j_name, similarity))
face_test.addSample(i_name, scores)
#print face_test.rank1_bounds
self.assertAlmostEqual(face_test.rank1_rate, 0.43930635838150289)
self.assertAlmostEqual(face_test.rank1_bounds[0], 0.3640772723094895)
self.assertAlmostEqual(face_test.rank1_bounds[1], 0.51665118592791259)
roc = face_test.getROCAnalysis()
# Test based of fpr=0.01
_ = roc.getFAR(far=0.01)
class _TestFacePCA(unittest.TestCase):
def setUp(self):
self.images = []
self.names = []
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
self.images.append(img)
self.names.append(filename)
self.assert_( len(self.images) == 173 )
def test_pca_scraps(self):
face_test = FaceRecognitionTest.FaceRecognitionTest(name='PCA_CSUScraps',score_type=FaceRecognitionTest.SCORE_TYPE_HIGH)
pca = PCA(drop_front=2,basis_vectors=55)
for im_name in self.eyes.files():
im = Image(os.path.join(SCRAPS_FACE_DATA, im_name + ".pgm"))
rect = self.eyes.getFaces(im_name)
eyes = self.eyes.getEyes(im_name)
pca.addTraining(im,rect=rect[0],eyes=eyes[0])
pca.train()
face_records = {}
for im_name in self.eyes.files():
im = Image(os.path.join(SCRAPS_FACE_DATA, im_name + ".pgm"))
rect = self.eyes.getFaces(im_name)
eyes = self.eyes.getEyes(im_name)
fr = pca.computeFaceRecord(im,rect=rect[0],eyes=eyes[0])
face_records[im_name] = fr
for i_name in face_records.keys():
scores = []
for j_name in face_records.keys():
similarity = pca.similarity(face_records[i_name],face_records[j_name])
scores.append((j_name,similarity))
face_test.addSample(i_name,scores)
#print face_test.rank1_bounds
self.assertAlmostEqual(face_test.rank1_rate,0.43930635838150289)
self.assertAlmostEqual(face_test.rank1_bounds[0],0.3640772723094895)
self.assertAlmostEqual(face_test.rank1_bounds[1],0.51665118592791259)
roc = face_test.getROCAnalysis()
# Test based of fpr=0.01
roc_point = roc.getFAR(far=0.01)
def test_face_detection_pickle(self):
fd = CascadeDetector(OPENCV_CASCADE)
fdt = FaceDetectionTest(name='scraps')
buffer = pickle.dumps(fd)
fd = pickle.loads(buffer)
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth,rects,im=img)
self.assertAlmostEqual( fdt.pos_rate , 0.98265895953757221, places = 2 ) # TODO: Version 2 performance is better
def test_training(self):
'''
This trains the FaceFinder on the scraps database.
'''
#import cProfile
# Load an eyes file
eyes_filename = join(pv.__path__[0], 'data', 'csuScrapShots',
'coords.txt')
#print "Creating eyes File."
eyes_file = EyesFile(eyes_filename)
# Create a face detector
cascade_file = join(pv.__path__[0], 'config',
'facedetector_celebdb2.xml')
#print "Creating a face detector from:",cascade_file
face_detector = CascadeDetector(cascade_file)
image_dir = join(pv.__path__[0], 'data', 'csuScrapShots')
ed = SVMEyeDetectorFromDatabase(eyes_file,
image_dir,
image_ext=".pgm",
face_detector=face_detector,
random_seed=0)
edt = EyeDetectionTest(name='scraps')
#print "Testing..."
for img in self.images:
#print img.filename
faces = ed.detect(img)
#faces = ed.detect(img)
pred_eyes = []
for _, _, pleye, preye in faces:
#detections.append(rect)
pred_eyes.append((pleye, preye))
truth_eyes = self.eyes.getEyes(img.filename)
edt.addSample(truth_eyes, pred_eyes, im=img, annotate=False)
class _TestCascadeDetector(unittest.TestCase):
''' Unit tests for the Cascade Detector '''
def test_detect_bad_file(self):
'''
If the cascade file does not exist, opencv can crash without warning.
This makes sure a test is run to make sure the cascade is there.
'''
self.assertRaises(CascadeNotFound, CascadeDetector, BAD_CASCADE)
def test_face_detection_pickle(self):
fd = CascadeDetector(OPENCV_CASCADE)
fdt = FaceDetectionTest(name='scraps')
buffer = pickle.dumps(fd)
fd = pickle.loads(buffer)
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA, "coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth, rects, im=img)
self.assertAlmostEqual(
fdt.pos_rate, 0.98265895953757221,
places=2) # TODO: Version 2 performance is better
def test_detect_scraps_opencv(self):
fd = CascadeDetector(OPENCV_CASCADE)
fdt = FaceDetectionTest(name='scraps')
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA, "coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth, rects, im=img)
self.assertAlmostEqual(
fdt.pos_rate, 0.98265895953757221,
places=2) # TODO: Version 2 performance is better
def test_detect_scraps_celeb1(self):
fd = CascadeDetector(CELEB1_CASCADE)
fdt = FaceDetectionTest(name='scraps')
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA, "coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth, rects, im=img)
self.assertAlmostEqual(
fdt.pos_rate, 0.76878612716763006,
places=2) # TODO: Version 2 performance is better
def test_detect_scraps_celeb2(self):
fd = CascadeDetector(CELEB2_CASCADE)
fdt = FaceDetectionTest(name='scraps')
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA, "coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth, rects, im=img)
self.assertAlmostEqual(fdt.pos_rate, 0.925, places=2)
def donttest_detector_train(
self): # TODO: Cascade training fails for Version OpenCV 2.0
positives = []
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA, "coords.txt"))
n = len(self.eyes.files())
for filename in self.eyes.files()[:n / 2]:
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
faces = self.eyes.getFaces(img.filename)
positives.append(
[os.path.join(SCRAPS_FACE_DATA, img.filename), faces])
neg_files = []
for im_name in os.listdir(NONFACE_DATA):
if im_name[-4:] != ".jpg": continue
neg_files.append(os.path.join(NONFACE_DATA, im_name))
fd = trainHaarClassifier(positives,
neg_files,
nstages=6,
maxtreesplits=0,
max_run_time=300)
fdt = FaceDetectionTest(name='scraps')
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA, "coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth, rects, im=img)
self.assertAlmostEqual(fdt.pos_rate, 0.9942196531791907)
class _FastFilterTest(unittest.TestCase):
def setUp(self):
SCRAPS_FACE_DATA = os.path.join(pv.__path__[0],"data","csuScrapShots")
self.test_images = []
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files()[0:10]:
im = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
eyes = self.eyes.getEyes(filename)
#print eyes
affine = pv.AffineFromPoints(eyes[0][0],eyes[0][1],pv.Point(40,40),pv.Point(88,40),(128,128))
im = affine.transformImage(im)
self.test_images.append(im)
def test_gabor1(self):
ilog = None # pv.ImageLog(name="GaborTest1")
bank = FilterBank(tile_size=(128,128))
kernels = createGaborKernels()
for wavelet in kernels:
bank.addFilter(wavelet)
for i in range(len(bank.filters)):
filter = np.fft.ifft2(bank.filters[i])
if ilog:
ilog.log(pv.Image(np.fft.fftshift(filter.real)),label="Filter_RE_%d"%i)
ilog.log(pv.Image(np.fft.fftshift(filter.imag)),label="Filter_IM_%d"%i)
for im in self.test_images[:1]:
if ilog:
ilog.log(im,label="ORIG")
results = bank.convolve(im)
#print "RShape",results.shape[2]
if ilog:
for i in range(results.shape[2]):
ilog.log(pv.Image(results[:,:,i].real),label="CONV_RE_%d"%i)
ilog.log(pv.Image(results[:,:,i].imag),label="CONV_IM_%d"%i)
if ilog:
ilog.show()
def test_gabor2(self):
#print
bank = FilterBank(tile_size=(128,128))
#print "generating filters"
freq = 5
oreint = 3
test_values = [[0.39182228, 0.39265844, 5.4866541e-07, 7.4505806e-08, -0.00035403497, -0.049227916], [0.39180198, 0.3926785, -1.0430813e-07, 4.4703484e-08, 0.017195048, -0.045993667], [0.39180198, 0.39267856, -5.9604645e-08, 7.4505806e-09, 0.045639627, 0.017549083], [0.1959009, 0.19633935, -1.0430813e-07, -1.4901161e-08, -0.054863166, -0.010506729], [0.19590084, 0.19633923, -1.0337681e-07, -3.7252903e-08, -0.050385918, -0.024042567], [0.19590083, 0.19633923, -1.8626451e-07, -2.6077032e-08, 0.053237237, 0.014138865], [0.097950377, 0.098169744, 7.5995922e-07, 4.0978193e-08, -0.029739097, 0.029439665], [0.097950377, 0.098169573, 8.5681677e-08, 7.4505806e-09, -0.034587309, 0.023616293], [0.097950369, 0.098169573, 3.7252903e-09, 2.7939677e-08, 0.040645007, 0.0075459499], [0.048975188, 0.049084734, -1.0859221e-06, 1.4528632e-07, -0.0026100441, 0.025389811], [0.048975196, 0.049084827, -9.3132257e-09, 3.3527613e-08, -0.0058529032, 0.02486741], [0.048975192, 0.049084831, -7.6368451e-08, -4.703179e-08, 0.025110571, 0.0032778508], [0.024487557, 0.024542304, -0.00027022697, 1.1050142e-06, 0.0053004427, 0.013041494], [0.024487549, 0.024542348, -4.4152141e-05, 3.0389987e-05, 0.0040912526, 0.013478963], [0.024487551, 0.024542345, -4.4191256e-05, 3.0397903e-05, 0.013955923, 0.001284558]]
start = time.time()
i = 0
for f in range(freq):
#f=f+2
for o in range(oreint):
#print "F:%10.5f O:%10.5f"%(f,o)
w1 = GaborWavelet(np.pi*2.0**(-(f+2.0)/2.0),o*np.pi/oreint,np.pi)
mask = w1.mask((64,64))
ssr = (np.real(mask)**2).sum()
ssi = (np.imag(mask)**2).sum()
sr = np.real(mask).sum()
si = np.imag(mask).sum()
self.assertAlmostEqual(test_values[i][0],ssr,places=5)
self.assertAlmostEqual(test_values[i][1],ssi,places=5)
self.assertAlmostEqual(test_values[i][2],sr,places=5)
self.assertAlmostEqual(test_values[i][3],si,places=5)
self.assertAlmostEqual(test_values[i][4],np.real(mask)[3,2],places=5)
self.assertAlmostEqual(test_values[i][5],np.imag(mask)[3,2],places=5)
bank.addFilter(w1)
i+=1
#mask = w1.mask((128,128))
#Image(real(mask)).show()
#Image(imag(mask)).show()
stop = time.time()
#print "Filter Time:",stop-start
start = time.time()
for im in self.test_images:
jets = bank.convolve(im)
#mag = sqrt(real(jets)*real(jets)+imag(jets)*imag(jets))
#if ilog: ilog.log(im)
#for i in range(freq*oreint):
# if ilog: ilog.log(Image(real(jets[:,:,i])))
# if ilog: ilog.log(Image(real(jets[:,:,i])))
# if ilog: ilog.log(Image(mag[:,:,i]))
stop = time.time()
frame_time = (stop-start)/len(self.test_images)
#print "Frame Time: %0.5fs"% frame_time
def test_GaborFilters(self):
ilog = None # pv.ImageLog(name="GaborTest1")
#bank = FilterBank(tile_size=(128,128))
kernels = createGaborKernels()
filters = GaborFilters(kernels)
gim = filters.convolve(self.test_images[0])
template = gim.extractJet(pv.Point(64,64))
table = pv.Table()
for i in range(0,128):
table[i,'disp'] = i - 64
novel = gim.extractJet(pv.Point(i,64))
table[i,'simMag'] = template.simMag(novel)
table[i,'simPhase'] = template.simPhase(novel)
table[i,'simDisplace'] = template.simDisplace(novel)
dx,dy = template.displace(novel,method="Simple")
table[i,'displace_dx'] = dx
table[i,'displace_dy'] = dy
dx,dy = template.displace(novel,method="Blocked")
table[i,'blocked_dx'] = dx
table[i,'blocked_dy'] = dy
dx,dy = template.displace(novel,method="Iter")
table[i,'iter_dx'] = dx
table[i,'iter_dy'] = dy
#print
#print table
#table.save("../../gabor_plot.csv")
def test_GaborImage(self):
ilog = None # pv.ImageLog(name="GaborTest1")
#bank = FilterBank(tile_size=(128,128))
kernels = createGaborKernels()
filters = GaborFilters(kernels)
gim = filters.convolve(self.test_images[0])
test_point = pv.Point(62.6,64.8)
template = gim.extractJet(test_point)
new_point,sim,jet = gim.locatePoint(template,pv.Point(60,70))
self.assertAlmostEqual(new_point.l2(test_point),0.0)
new_point,sim,jet = gim.locatePoint(template,pv.Point(30,49))
self.assert_(new_point.l2(test_point) > 1.0)
new_point,sim,jet = gim.locatePoint(template)
self.assertAlmostEqual(new_point.l2(test_point),0.0)
class _TestCascadeDetector(unittest.TestCase):
''' Unit tests for the Cascade Detector '''
def test_detect_bad_file(self):
'''
If the cascade file does not exist, opencv can crash without warning.
This makes sure a test is run to make sure the cascade is there.
'''
self.assertRaises(CascadeNotFound,CascadeDetector,BAD_CASCADE)
def test_face_detection_pickle(self):
fd = CascadeDetector(OPENCV_CASCADE)
fdt = FaceDetectionTest(name='scraps')
buffer = pickle.dumps(fd)
fd = pickle.loads(buffer)
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth,rects,im=img)
self.assertAlmostEqual( fdt.pos_rate , 0.98265895953757221, places = 2 ) # TODO: Version 2 performance is better
def test_detect_scraps_opencv(self):
fd = CascadeDetector(OPENCV_CASCADE)
fdt = FaceDetectionTest(name='scraps')
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth,rects,im=img)
self.assertAlmostEqual( fdt.pos_rate , 0.98265895953757221, places = 2 ) # TODO: Version 2 performance is better
def test_detect_scraps_celeb1(self):
fd = CascadeDetector(CELEB1_CASCADE)
fdt = FaceDetectionTest(name='scraps')
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth,rects,im=img)
self.assertAlmostEqual( fdt.pos_rate , 0.76878612716763006, places = 2 ) # TODO: Version 2 performance is better
def test_detect_scraps_celeb2(self):
fd = CascadeDetector(CELEB2_CASCADE)
fdt = FaceDetectionTest(name='scraps')
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth,rects,im=img)
self.assertAlmostEqual( fdt.pos_rate , 0.925, places = 2 )
def donttest_detector_train(self): # TODO: Cascade training fails for Version OpenCV 2.0
positives = []
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
n = len(self.eyes.files())
for filename in self.eyes.files()[:n/2]:
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
faces = self.eyes.getFaces(img.filename)
positives.append([os.path.join(SCRAPS_FACE_DATA,img.filename),faces])
neg_files = []
for im_name in os.listdir(NONFACE_DATA):
if im_name[-4:] != ".jpg": continue
neg_files.append(os.path.join(NONFACE_DATA,im_name))
fd = trainHaarClassifier(positives,neg_files,nstages=6,maxtreesplits=0,max_run_time=300)
fdt = FaceDetectionTest(name='scraps')
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files():
img = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
rects = fd(img)
truth = self.eyes.getFaces(img.filename)
fdt.addSample(truth,rects,im=img)
self.assertAlmostEqual( fdt.pos_rate , 0.9942196531791907 )
class _FastFilterTest(unittest.TestCase):
def setUp(self):
SCRAPS_FACE_DATA = os.path.join(pv.__path__[0],"data","csuScrapShots")
self.test_images = []
self.eyes = EyesFile(os.path.join(SCRAPS_FACE_DATA,"coords.txt"))
for filename in self.eyes.files()[0:10]:
im = pv.Image(os.path.join(SCRAPS_FACE_DATA, filename + ".pgm"))
eyes = self.eyes.getEyes(filename)
#print eyes
affine = pv.AffineFromPoints(eyes[0][0],eyes[0][1],pv.Point(40,40),pv.Point(88,40),(128,128))
im = affine.transformImage(im)
self.test_images.append(im)
def test_gabor1(self):
ilog = None # pv.ImageLog(name="GaborTest1")
bank = FilterBank(tile_size=(128,128))
kernels = createGaborKernels()
for wavelet in kernels:
bank.addFilter(wavelet)
for i in range(len(bank.filters)):
corr_filter = np.fft.ifft2(bank.filters[i])
if ilog:
ilog.log(pv.Image(np.fft.fftshift(corr_filter.real)),label="Filter_RE_%d"%i)
ilog.log(pv.Image(np.fft.fftshift(corr_filter.imag)),label="Filter_IM_%d"%i)
for im in self.test_images[:1]:
if ilog:
ilog.log(im,label="ORIG")
results = bank.convolve(im)
#print "RShape",results.shape[2]
if ilog:
for i in range(results.shape[2]):
ilog.log(pv.Image(results[:,:,i].real),label="CONV_RE_%d"%i)
ilog.log(pv.Image(results[:,:,i].imag),label="CONV_IM_%d"%i)
if ilog:
ilog.show()
def test_gabor2(self):
#print
bank = FilterBank(tile_size=(128,128))
#print "generating filters"
freq = 5
oreint = 3
test_values = [[0.39182228, 0.39265844, 5.4866541e-07, 7.4505806e-08, -0.00035403497, -0.049227916], [0.39180198, 0.3926785, -1.0430813e-07, 4.4703484e-08, 0.017195048, -0.045993667], [0.39180198, 0.39267856, -5.9604645e-08, 7.4505806e-09, 0.045639627, 0.017549083], [0.1959009, 0.19633935, -1.0430813e-07, -1.4901161e-08, -0.054863166, -0.010506729], [0.19590084, 0.19633923, -1.0337681e-07, -3.7252903e-08, -0.050385918, -0.024042567], [0.19590083, 0.19633923, -1.8626451e-07, -2.6077032e-08, 0.053237237, 0.014138865], [0.097950377, 0.098169744, 7.5995922e-07, 4.0978193e-08, -0.029739097, 0.029439665], [0.097950377, 0.098169573, 8.5681677e-08, 7.4505806e-09, -0.034587309, 0.023616293], [0.097950369, 0.098169573, 3.7252903e-09, 2.7939677e-08, 0.040645007, 0.0075459499], [0.048975188, 0.049084734, -1.0859221e-06, 1.4528632e-07, -0.0026100441, 0.025389811], [0.048975196, 0.049084827, -9.3132257e-09, 3.3527613e-08, -0.0058529032, 0.02486741], [0.048975192, 0.049084831, -7.6368451e-08, -4.703179e-08, 0.025110571, 0.0032778508], [0.024487557, 0.024542304, -0.00027022697, 1.1050142e-06, 0.0053004427, 0.013041494], [0.024487549, 0.024542348, -4.4152141e-05, 3.0389987e-05, 0.0040912526, 0.013478963], [0.024487551, 0.024542345, -4.4191256e-05, 3.0397903e-05, 0.013955923, 0.001284558]]
i = 0
for f in range(freq):
#f=f+2
for o in range(oreint):
#print "F:%10.5f O:%10.5f"%(f,o)
w1 = GaborWavelet(np.pi*2.0**(-(f+2.0)/2.0),o*np.pi/oreint,np.pi)
mask = w1.mask((64,64))
ssr = (np.real(mask)**2).sum()
ssi = (np.imag(mask)**2).sum()
sr = np.real(mask).sum()
si = np.imag(mask).sum()
self.assertAlmostEqual(test_values[i][0],ssr,places=5)
self.assertAlmostEqual(test_values[i][1],ssi,places=5)
self.assertAlmostEqual(test_values[i][2],sr,places=5)
self.assertAlmostEqual(test_values[i][3],si,places=5)
self.assertAlmostEqual(test_values[i][4],np.real(mask)[3,2],places=5)
self.assertAlmostEqual(test_values[i][5],np.imag(mask)[3,2],places=5)
bank.addFilter(w1)
i+=1
#mask = w1.mask((128,128))
#Image(real(mask)).show()
#Image(imag(mask)).show()
#print "Filter Time:",stop-start
def test_GaborFilters(self):
#bank = FilterBank(tile_size=(128,128))
kernels = createGaborKernels()
filters = GaborFilters(kernels)
gim = filters.convolve(self.test_images[0])
template = gim.extractJet(pv.Point(64,64))
table = pv.Table()
for i in range(0,128):
table[i,'disp'] = i - 64
novel = gim.extractJet(pv.Point(i,64))
table[i,'simMag'] = template.simMag(novel)
table[i,'simPhase'] = template.simPhase(novel)
table[i,'simDisplace'] = template.simDisplace(novel)
dx,dy = template.displace(novel,method="Simple")
table[i,'displace_dx'] = dx
table[i,'displace_dy'] = dy
dx,dy = template.displace(novel,method="Blocked")
table[i,'blocked_dx'] = dx
table[i,'blocked_dy'] = dy
dx,dy = template.displace(novel,method="Iter")
table[i,'iter_dx'] = dx
table[i,'iter_dy'] = dy
#print
#print table
#table.save("../../gabor_plot.csv")
def test_GaborImage(self):
kernels = createGaborKernels()
filters = GaborFilters(kernels)
gim = filters.convolve(self.test_images[0])
test_point = pv.Point(62.6,64.8)
template = gim.extractJet(test_point)
new_point,_,_ = gim.locatePoint(template,pv.Point(60,70))
self.assertAlmostEqual(new_point.l2(test_point),0.0)
new_point,_,_ = gim.locatePoint(template,pv.Point(30,49))
self.assert_(new_point.l2(test_point) > 1.0)
new_point,_,_ = gim.locatePoint(template)
self.assertAlmostEqual(new_point.l2(test_point),0.0)
