def getPhowFeatures(imagedata, phowOpts):
im = standardizeImage(imagedata)
frames, descrs = vl_phow(im,
verbose=phowOpts.Verbose,
sizes=phowOpts.Sizes,
step=phowOpts.Step)
return frames, descrs
def getPhowFeatures(imagedata, phowOpts):
im = standarizeImage(imagedata)
frames, descrs = vl_phow(im,
verbose=phowOpts.Verbose,
sizes=phowOpts.Sizes,
step=phowOpts.Step)
return frames, descrs
def getImageDescriptor(model, im):
import numpy as np
from vl_phow import vl_phow
im = standarizeImage(im, resize=200)
width = im.shape[1]
height = im.shape[0]
numWords = model.vocab.shape[1]
model.phowOpts['verbose'] = False
# get PHOW features
frames, descrs = vl_phow(im, **model.phowOpts)
# quantize local descriptors into visual words
if model.quantizer == 'kdtree':
binsa = model.kdtree.query(descrs.T.astype(np.float32))[1]
hists = []
for i in range(len(model.numSpatialX)):
binsx = np.digitize(
frames[1], np.linspace(0, width, model.numSpatialX[i] + 1)) - 1
binsy = np.digitize(
frames[0], np.linspace(0, height, model.numSpatialY[i] + 1)) - 1
# combined quantization
bins = np.ravel_multi_index(
(binsy, binsx, binsa),
[model.numSpatialY[i], model.numSpatialX[i], numWords])
hist = np.histogram(bins, len(bins))[0].astype(np.float32)
hists.append(hist / np.sum(hist))
hist = np.concatenate(hists, axis=0)
hist = hist / np.sum(hist)
return hist
def getPhowFeatures(imagedata, phowOpts): #extracts features from image im = standardizeImage(imagedata) #scale image to 640x480 frames, descrs = vl_phow(im, verbose=phowOpts.Verbose, sizes=phowOpts.Sizes, step=phowOpts.Step) return frames, descrs
#!/usr/bin/python
from vl_phow import vl_phow
from numpy import *
from scipy.io import savemat, loadmat
I = loadmat('lena.mat')['lena_gray']
I = asfortranarray(I, dtype=float32)
f, d = vl_phow(I, step=50, floatDescriptors=True, verbose=True)
savetxt('phow_f_python.txt', f.T, fmt='%.2f')
savetxt('phow_d_python.txt', d.T, fmt='%.2f')
model.b = []
model.classify = classify
# --------------------------------------------------------------------
# Train vocabulary
# --------------------------------------------------------------------
if not os.path.isfile(conf.vocabPath):
# Get some PHOW descriptors to train the dictionary
selTrainFeats = random_subset(selTrain, 30)
descrs = []
for ii in range(len(selTrainFeats)):
im = imageio.imread(images[selTrainFeats[ii]])
im = standarizeImage(im)
descrs.append(vl_phow(im, **model.phowOpts)[1])
descrs = np.concatenate(descrs, axis=1).T
descrs = random_subset(descrs.tolist(), int(40e4))
descrs = np.array(descrs).T.astype(np.float32)
# Quantize the descriptors to get the visual words
vocab = kmeans(descrs.T.copy(order='C'),
conf.numWords,
verbose=True,
algorithm='ELKAN',
max_num_iterations=100)
# Required .copy(order='C') because of cython
vocab = vocab.T
save(conf.vocabPath, vocab)
else:
