def __init__(self, cam, **kwargs):
self.nodes = set()
self.edges = set()
self.every_edge = []
self.weak_edges = []
self.link_thresh = 100 # how many inliers needed to add in a long-range link
if 1:
self.pg = TreeOptimizer3()
self.vset = set()
self.oldvset = set()
else:
import faketoro
self.pg = faketoro.faketoro()
self.pg.initializeOnlineOptimization()
self.place_ids = []
self.cam = cam
self.pe = PoseEstimator(*cam.params)
self.optimize_after_addition = True
self.ds = None
for k,a in kwargs.items():
if k == 'descriptor_scheme':
self.ds = a
elif k == 'optimize_after_addition':
self.optimize_after_addition = a
elif k == 'link_thresh':
self.link_thresh = a
if self.ds == None:
self.ds = DescriptorSchemeCalonder()
self.vt = None
search = [ '/u/mihelich/images/holidays/holidays.tree', '/u/jamesb/holidays.tree' ]
for filename in search:
if os.access(filename, os.R_OK):
self.vt = place_recognition.load(filename, self.ds.cl)
break
if not self.vt:
print "ERROR: Could not find a valid place_recognition tree in", search
assert 0
self.node_kp = {}
self.node_descriptors = {}
self.termcrit = default_termcrit
self.pr_maximum = 15 # How many out of PR's places to consider for GCC
self.node_vdist = 15 # how many frame to wait to put in a skeleton node
self.adaptive = False
self.label = "no label"
self.node_labels = {}
self.fills = False
self.timer = {}
for t in ['toro add', 'toro opt', 'place recognition', 'gcc', 'descriptors']:
self.timer[t] = Timer()
def __init__(self, ds):
self.cameras = []
self.pics = []
filename = '/u/mihelich/images/holidays/holidays.tree'
self.vt = place_recognition.load(filename)
self.pe = PoseEstimator()
self.clear()
self.ds = ds
self.verbose = 1
def match_runs(reference_files, query_files):
print "loading vocabulary tree..."
vt = place_recognition.load("/u/mihelich/images/holidays/holidays.tree")
cl = calonder.classifier()
cl.read('/u/prdata/calonder_trees/current.rtc')
print "adding database images..."
for i in reference_files[2]:
# Add image to vocabulary tree
name = reference_files[0] % i
img = Image.open(name)
desc, kp = get_features(img, cl)
vt.add(img, desc)
print '%s has %d feature points' % (name, len(kp))
# Save coordinates and descriptors
file = open(reference_files[1] % (i, "key"), "w")
for (pt, d) in zip(kp, desc):
# Format: x y [d...]
file.write("%i %i %s\n" % (pt[0], pt[1], str(list(d))))
print "getting top-k views for query images..."
N = len(reference_files[2])
k = 32
M = []
for i in query_files[2]:
# Get indexes of best N matching views from reference run
name = query_files[0] % i
img = Image.open(name)
desc, kp = get_features(img, cl)
print '%s has %d feature points' % (name, len(kp))
scores = vt.topN(img, desc, N)
M.append(scores)
top_n = sorted(enumerate(scores), key=operator.itemgetter(1))[:k]
# Save filenames of matching views
file = open(query_files[1] % (i, "match"), "w")
for (index, score) in top_n:
file.write("%s %f\n" %
(reference_files[0] % reference_files[2][index], score))
return M
import pylab, numpy
import Image
import sys
import random
sys.path.append('lib')
import place_recognition
#import calonder
#import fast
ims = [
Image.open("/u/prdata/videre-bags/james4/im.%06u.left_rectified.tiff" %
(20 * i)) for i in range(100)
]
random.seed(0)
BUILD_TREE = True
if BUILD_TREE:
vt = place_recognition.vocabularytree()
vt.build(random.sample(ims, 50), 5, 4, False)
vt.save("thrash.tree")
else:
vt = place_recognition.load("thrash.tree")
for (a, q) in zip(random.sample(ims, 100), random.sample(ims, 100)):
print a, q
vt.add(a)
vt.topN(q, None, 10)
print 2, newpose(2).xform(0, 0, 0)
sys.exit(1)
if 0:
vt = None
else:
if 0:
vt = place_recognition.vocabularytree()
ims = [
Image.open(
"/u/prdata/videre-bags/james4/im.%06u.left_rectified.tiff" %
(200 * i)) for i in range(10)
]
vt.build(ims, 5, 4, False)
else:
vt = place_recognition.load(
"/u/mihelich/images/holidays/holidays.tree")
vo = VisualOdometer(cam,
scavenge=False,
feature_detector=FeatureDetectorFast(),
descriptor_scheme=DescriptorSchemeCalonder(),
inlier_error_threshold=3.0,
sba=None,
inlier_thresh=99999,
position_keypoint_thresh=0.2,
angle_keypoint_thresh=0.15)
from skeleton import Skeleton
if 0:
skel = Skeleton(cam)
skel.node_vdist = 0
outlet_files = '/wg/wgdata1/vol1/Outlets/OutletWhiteHall_cropped_outlets/IMG_%04d.JPG'
no_out_files = '/wg/wgdata1/vol1/Outlets/OutletWhiteHall_no_outlet/IMG_%04d.JPG'
print 'opening images...'
outlet_train = [
Image.open(outlet_files % i).convert('L') for i in range(416, 427)
]
no_out_train = [
Image.open(no_out_files % i).convert('L') for i in range(416, 427)
]
#outlet_test = [ Image.open(outlet_files % i).convert('L') for i in range(372, 416) + range(427, 471) ]
#no_out_test = [ Image.open(no_out_files % i).convert('L') for i in range(372, 416) + range(427, 471) ]
print 'loading trees...'
vt = place_recognition.load('/u/mihelich/images/holidays/holidays_b5d4.tree')
#vt = place_recognition.load('/u/mihelich/images/holidays/holidays.tree')
#vt = place_recognition.vocabularytree()
#vt.build(outlet_train, 5, 4, False)
cl = calonder.classifier()
cl.read('/u/prdata/calonder_trees/current.rtc')
def kp_d(frame):
fkp = fast.fast(frame.tostring(), frame.size[0], frame.size[1], 10, 15)
fkp = [(x, y, r) for (x, y, r) in fkp
if (16 <= x and 16 <= y and x <= frame.size[0] -
16 and y <= frame.size[1] - 16)]
fkp = fast.nonmax(fkp) #damn this is slow
kp = [(x, y) for (
x, y,