def vl_test_ikmeans():
# VL_TEST_IKMEANS Test VL_IKMEANS function
print "test_ikmeans: Testing VL_IKMEANS and IKMEANSPUSH"
# -----------------------------------------------------------------------
print "test_ikmeans: Testing Lloyd algorithm"
K = 5
data = numpy.array(numpy.random.rand(2, 1000) * 255, "uint8")
datat = numpy.array(numpy.random.rand(2, 10000) * 255, "uint8")
[C, A] = vlfeat.vl_ikmeans(data, K, verbose=1)
AT = vlfeat.vl_ikmeanspush(datat, C, verbose=1)
plot_partition(data, datat, C, A, AT)
pylab.title("vl_ikmeans (Lloyd algorithm)")
pylab.xlim(0, 255)
pylab.ylim(0, 255)
print "ikmeans_lloyd"
pylab.figure()
[C, A] = vlfeat.vl_ikmeans(data, K, verbose=1, method="elkan")
AT = vlfeat.vl_ikmeanspush(datat, C, verbose=1, method="elkan")
pylab.title("vl_ikmeans (Elkan algorithm)")
plot_partition(data, datat, C, A, AT)
pylab.xlim(0, 255)
pylab.ylim(0, 255)
print "ikmeans_elkan"
pylab.show()
def vl_test_ikmeans():
# VL_TEST_IKMEANS Test VL_IKMEANS function
print ('test_ikmeans: Testing VL_IKMEANS and IKMEANSPUSH')
# -----------------------------------------------------------------------
print ('test_ikmeans: Testing Lloyd algorithm')
K = 5
data = numpy.array(numpy.random.rand(2,1000) * 255, 'uint8')
datat = numpy.array(numpy.random.rand(2,10000)* 255, 'uint8')
[C, A] = vlfeat.vl_ikmeans(data, K, verbose=1)
AT = vlfeat.vl_ikmeanspush(datat, C, verbose=1)
plot_partition(data, datat, C, A, AT)
pylab.title('vl_ikmeans (Lloyd algorithm)')
pylab.xlim(0, 255)
pylab.ylim(0, 255)
print ('ikmeans_lloyd')
pylab.figure()
[C, A] = vlfeat.vl_ikmeans(data, K, verbose=1, method='elkan')
AT = vlfeat.vl_ikmeanspush(datat, C, verbose=1, method='elkan')
pylab.title('vl_ikmeans (Elkan algorithm)')
plot_partition(data, datat, C, A, AT)
pylab.xlim(0, 255)
pylab.ylim(0, 255)
print ('ikmeans_elkan')
pylab.show()
def trainVocab(selTrain, all_images, conf):
selTrainFeats = sample(selTrain, conf.images_for_histogram)
descrs = []
if MULTIPROCESSING:
raise ValueError('MULTIPROCESSING not implemented')
#pool = Pool(processes=30)
#list_of_train_images = [all_images[i] for i in selTrainFeats]
#descrs.append(pool.map_async(getPhowFeatures, list_of_train_images).get())
else:
for i in selTrainFeats:
im = imread(all_images[i])
# Debugging
#print all_images[i], 'shape', im.shape
descrs.append(getPhowFeatures(im, conf.phowOpts)[1])
# the '[1]' is there because we only want the descriptors and not the frames
descrs = hstack(descrs)
n_features = descrs.shape[1]
sample_indices = sample(arange(n_features),
conf.numbers_of_features_for_histogram)
descrs = descrs[:, sample_indices]
descrs = array(descrs, 'uint8')
# Quantize the descriptors to get the visual words
vocab, _ = vl_ikmeans(descrs,
K=conf.numWords,
verbose=conf.verbose,
method='elkan')
return vocab
def trainVocab(selTrain, all_images, conf):
selTrainFeats = sample(selTrain, conf.images_for_histogram)
descrs = []
if MULTIPROCESSING:
raise ValueError('MULTIPROCESSING not implemented')
#pool = Pool(processes=30)
#list_of_train_images = [all_images[i] for i in selTrainFeats]
#descrs.append(pool.map_async(getPhowFeatures, list_of_train_images).get())
else:
for i in selTrainFeats:
im = imread(all_images[i])
descrs.append(getPhowFeatures(im, conf.phowOpts)[1])
# the '[1]' is there because we only want the descriptors and not the frames
descrs = hstack(descrs)
n_features = descrs.shape[1]
sample_indices = sample(arange(n_features), conf.numbers_of_features_for_histogram)
descrs = descrs[:, sample_indices]
descrs = array(descrs, 'uint8')
# Quantize the descriptors to get the visual words
vocab, _ = vl_ikmeans(descrs,
K=conf.numWords,
verbose=conf.verbose,
method='elkan')
return vocab
def trainVocab(selTrain, all_images, conf): selTrainFeats = sample(selTrain, conf.images_for_histogram) descrs = [] #start multiprocessing block pool = multiprocessing.Pool(processes=conf.numCore) results = [pool.apply_async(getPhowFeaturesMulti, args=(imread(all_images[i]), conf.phowOpts, i)) for i in selTrainFeats] descrs = [p.get() for p in results] sorted(descrs) for descr in descrs: descr.pop(0) depict = [] for descr in descrs: depict.append(descr[0]) descrs = depict #end multiprocessing block descrs = hstack(descrs) n_features = descrs.shape[1] sample_indices = sample(arange(n_features), conf.numbers_of_features_for_histogram) descrs = descrs[:, sample_indices] descrs = array(descrs, 'uint8') # Quantize the descriptors to get the visual words vocab, _ = vl_ikmeans(descrs, K=conf.numWords, verbose=conf.verbose, method='elkan') return vocab
def trainVocab(selTrain, all_images, conf): selTrainFeats = sample(selTrain, conf.images_for_histogram) descrs = [] #start multiprocessing block pool = multiprocessing.Pool(processes=conf.numCore) results = [pool.apply_async(getPhowFeaturesMulti, args=(imread(all_images[ii]), conf.phowOpts, i)) for i, ii in enumerate(selTrainFeats)] descrs = [p.get() for p in results] pool.terminate() sorted(descrs) for descr in descrs: descr.pop(0) depict = [] for descr in descrs: depict.append(descr[0]) descrs = depict #end multiprocessing block descrs = hstack(descrs) n_features = descrs.shape[1] sample_indices = sample(arange(n_features), conf.numbers_of_features_for_histogram) descrs = descrs[:, sample_indices] descrs = array(descrs, 'uint8') # Quantize the descriptors to get the visual words vocab, _ = vl_ikmeans(descrs, K=conf.numWords, verbose=conf.verbose, method='elkan') return vocab
def trainVocab(selTrain, all_images, conf):
selTrainFeats = sample(selTrain, conf.images_for_histogram)
descrs = []
for i in selTrainFeats:
im = imread(all_images[i])
descrs.append(getPhowFeatures(im, conf.phowOpts)[1])
# the '[1]' is there because we only want the descriptors and not the frames
descrs = hstack(descrs)
n_features = descrs.shape[1]
sample_indices = sample(arange(n_features), conf.numbers_of_features_for_histogram)
descrs = descrs[:, sample_indices]
descrs = array(descrs, 'uint8')
# Quantize the descriptors to get the visual words
vocab, _ = vl_ikmeans(descrs,
K=conf.numWords,
verbose=conf.verbose,
method='elkan')
return vocab
def learn(bk, descr):
if bk['dictionary'] == 'ikm':
print('block_dictionary: === Running IKM ===')
print('block_dictionary: num words (K): %d'%(bk['ikm_nwords']))
dict_ = vlfeat.vl_ikmeans(descr,
bk['ikm_nwords'],
verbosity=1,
method='elkan')
print('block_dictionary: IKM done')
elif bk['dictionary'] == 'hikm':
print('block_dictionary: === Runnining HIKM ===')
print('block_dictionary: num leaves: %d'%(bk['hikm_nleaves']))
print('block_dictionary: branching (K): %d'%(bk['hikm_K']))
print('block_dictionary: only_leaves: %d'%(bk['hikm_only_leaves']))
dict_,asgn = vlfeat.vl_hikmeans(descr,
bk['hikm_K'],
bk['hikm_nleaves'],
verbosity=1 ,
method='elkan')
print('block_dictionary: HIKM done')
else:
raise TypeError('block_dictionary.py :learn() unknown dictionary type')
return dict_
def cluster(self):
self.vector = self.vector.transpose()
centers, _ = vl_ikmeans(self.vector, self.clusters_count,method='elkan', verbose=0)
return centers.transpose()
