def __init__(self, feats=Feats()):
self.feats = feats
self.le = preprocessing.LabelEncoder()
self.cls = struct_perceptron.StructuredPerceptron(self,
max_iter=25,
average=True,
verbose=True)
def single_match(self, fid, sid, thres=0.9):
"""
BruteForce-L1 BruteForce-Hamming BruteForce-Hamming(2) FlannBased
"""
mod = dict(kp_algo=self.algorithm['keypoint'],
des_algo=self.algorithm['descriptor'])
mtype = self.algorithm['matching']
fimg, simg = self.__image_pair(fid, sid)
ff = Feats(self.root, self.name)
sif = ff.feats_set([simg], mod=mod)[0]
vif = ff.feats_set([fimg], mod=mod)[0]
mat = cv2.DescriptorMatcher_create(mtype)
mra = mat.knnMatch(sif['des'], vif['des'], k=2)
mra = self._filter_matches(mra, thres)
col = ['qix', 'tix', 'iix', 'dt']
if len(mra) is 0:
res = pd.DataFrame(columns=col)
else:
mra = self._pickle_matches(mra)
res = pd.DataFrame(mra, columns=col)
return dict(matches=res, sif=sif, vif=vif, qsize=simg['img'].shape)
def set_match(self, fids=[], ransac=0, thres=.9):
"""
ransac:
0, no ransac
1, use ransac
"""
mod = dict(kp_algo=self.algorithm['keypoint'],
des_algo=self.algorithm['descriptor'])
mtype = self.algorithm['matching']
sr = self.root
sn = self.name
ff = Feats(sr, sn)
isl = ff.o_slides(gray=True, resize=True)
vv = Video(sr, sn)
ivl = vv.get_frames(ids=fids, gray=True)
sfs = ff.feats_set(isl, mod=mod)
vfs = ff.feats_set(ivl, mod=mod)
res = []
for qf in vfs:
frs = []
for sf in sfs:
_info = self._match_info(sf['idx'], qf['idx'], mod['kp_algo'],
mod['des_algo'], mtype, thres)
self.elog.info(_info)
with ExpTimer(verbose=0) as ts:
mr = self._match_desc(sf['des'], qf['des'], mtype, thres)
if ransac is True and len(mr) > 0:
ra = Ransac()
h**o, mask = ra.compute(mr, sf['kps'], qf['kps'])
mr = mr[mask[mr.index] == 1]
mt = ts.msecs + sf['ts'] + qf['ts']
dr = dict(sid=sf['idx'], fid=qf['idx'], ts=mt, mr=mr)
frs.append(dr)
ds = self._dist_stat(frs)
vr = self._voting(ds)
res.append(dict(feats=frs, vr=vr))
return res, sfs, vfs
def __init__(self, epochs=10, batch_size=8, feats=Feats()):
self.feats = feats
from sklearn import preprocessing
self.le = preprocessing.LabelEncoder()
self.cls = struct_perceptron.StructuredPerceptron(
self,
max_iter=epochs,
batch_size=batch_size,
average=True,
verbose=True)
# max_len is set to be longer than any sentences in trainset
# if you want to use smaller number, you may need to need to spend some time debugging
self.max_len = 40
self.batch_size = batch_size
def __init__(self, feats=Feats()):
self.feats = feats
from sklearn.linear_model import LogisticRegression
self.cls = LogisticRegression()
from sklearn import preprocessing
self.le = preprocessing.LabelEncoder()
def __init__(self, feats=Feats()):
self.feats = feats
self.cls = LogisticRegression()
self.le = preprocessing.LabelEncoder()
def __init__(self, feats = Feats()):
self.feats = feats
self.le = preprocessing.LabelEncoder()
self.cls = struct_perceptron.StructuredPerceptron(self, max_iter=24, average=True, verbose=True)#, decay_t0=-0.05, decay_exponent=1)