def get_fps(detector, i, fileobj, size):
ag.info('{0} Initial processing {1}'.format(i, fileobj.img_id))
gen = generate_random_patches([fileobj.path], size, per_image=PER_IMAGE)
neg_feats = []
radii = detector.settings['spread_radii']
psize = detector.settings['subsample_size']
rotspread = detector.settings.get('rotation_spreading_radius', 0)
cb = detector.settings.get('crop_border')
setts = dict(spread_radii=radii, subsample_size=psize, rotation_spreading_radius=rotspread, crop_border=cb)
for neg in itr.islice(gen, PER_IMAGE):
#images.append(neg)
feat = detector.descriptor.extract_features(neg, settings=setts)
neg_feats.append(feat)
return neg_feats
import os.path
import glob
import numpy as np
import amitgroup as ag
import itertools
from scipy.stats import mstats
from train_superimposed import generate_random_patches
from settings import load_settings
settings = load_settings(settings_file)
descriptor = gv.load_descriptor(settings)
path = os.path.expandvars(settings['detector']['neg_dir'])
files = sorted(glob.glob(path))
gen = generate_random_patches(files, (100, 100), seed=0, per_image=5)
cut = 4
radii = settings['detector']['spread_radii']
psize = settings['detector']['subsample_size']
mm = []
for i, im in itertools.islice(enumerate(gen), limit):
print 'Processing {}'.format(i)
feats = descriptor.extract_features(im, settings=dict(spread_radii=radii, subsample_size=psize, crop_border=cut))
m = feats.mean()
mm.append(m)
import matplotlib.pylab as plt