def compute_weighte_average_score(match):
""" old scoring measure """
import vtool as vt
# Get distinctivness and forground of matching points
fx1_list, fx2_list = match.fm.T
annot1 = match.annot1
annot2 = match.annot2
dstncvs1 = annot1.dstncvs.take(fx1_list)
dstncvs2 = annot2.dstncvs.take(fx2_list)
fgweight1 = annot1.fgweights.take(fx1_list)
fgweight2 = annot2.fgweights.take(fx2_list)
dstncvs = np.sqrt(dstncvs1 * dstncvs2)
fgweight = np.sqrt(fgweight1 * fgweight2)
fsv = np.vstack((match.fs, dstncvs, fgweight)).T
fs_new = vt.weighted_average_scoring(fsv, [0], [1, 2])
weight_ave_score = fs_new.sum()
return weight_ave_score
def compute_weighte_average_score(match):
""" old scoring measure """
import vtool as vt
# Get distinctivness and forground of matching points
fx1_list, fx2_list = match.fm.T
annot1 = match.annot1
annot2 = match.annot2
dstncvs1 = annot1.dstncvs.take(fx1_list)
dstncvs2 = annot2.dstncvs.take(fx2_list)
fgweight1 = annot1.fgweights.take(fx1_list)
fgweight2 = annot2.fgweights.take(fx2_list)
dstncvs = np.sqrt(dstncvs1 * dstncvs2)
fgweight = np.sqrt(fgweight1 * fgweight2)
fsv = np.vstack((match.fs, dstncvs, fgweight)).T
fs_new = vt.weighted_average_scoring(fsv, [0], [1, 2])
weight_ave_score = fs_new.sum()
return weight_ave_score
def apply_new_qres_filter_scores(qreq_vsone_, qres_vsone, newfsv_list,
newscore_aids, filtkey):
r"""
applies the new filter scores vectors to a query result and updates other
scores
Args:
qres_vsone (QueryResult): object of feature correspondences and scores
newfsv_list (list):
newscore_aids (?):
filtkey (?):
CommandLine:
python -m ibeis.algo.hots.special_query --test-apply_new_qres_filter_scores
Example:
>>> # DISABLE_DOCTEST
>>> from ibeis.algo.hots.special_query import * # NOQA
>>> ibs, valid_aids = testdata_special_query()
>>> qaids = valid_aids[0:1]
>>> daids = valid_aids[1:]
>>> qaid = qaids[0]
>>> filtkey = hstypes.FiltKeys.DISTINCTIVENESS
>>> use_cache = False
>>> qaid2_qres_vsmany, qreq_vsmany_ = query_vsmany_initial(ibs, qaids, daids, use_cache)
>>> vsone_query_pairs = build_vsone_shortlist(ibs, qaid2_qres_vsmany)
>>> qaid2_qres_vsone, qreq_vsone_ = query_vsone_pairs(ibs, vsone_query_pairs, use_cache)
>>> qreq_vsone_.load_score_normalizer()
>>> qres_vsone = qaid2_qres_vsone[qaid]
>>> qres_vsmany = qaid2_qres_vsmany[qaid]
>>> top_aids = vsone_query_pairs[0][1]
>>> newfsv_list, newscore_aids = get_new_qres_distinctiveness(qres_vsone, qres_vsmany, top_aids, filtkey)
>>> apply_new_qres_filter_scores(qreq_vsone_, qres_vsone, newfsv_list, newscore_aids, filtkey)
Ignore:
qres_vsone.show_top(ibs, name_scoring=True)
print(qres_vsone.get_inspect_str(ibs=ibs, name_scoring=True))
print(qres_vsmany.get_inspect_str(ibs=ibs, name_scoring=True))
"""
assert ut.listfind(qres_vsone.filtkey_list, filtkey) is None
# HACK to update result cfgstr
qres_vsone.filtkey_list.append(filtkey)
qres_vsone.cfgstr = qreq_vsone_.get_cfgstr()
# Find positions of weight filters and score filters
# so we can apply a weighted average
#numer_filters = [hstypes.FiltKeys.LNBNN, hstypes.FiltKeys.RATIO]
weight_filters = hstypes.WEIGHT_FILTERS
weight_filtxs, nonweight_filtxs = vt.index_partition(
qres_vsone.filtkey_list, weight_filters)
for new_fsv_vsone, daid in zip(newfsv_list, newscore_aids):
#scorex_vsone = ut.listfind(qres_vsone.filtkey_list, filtkey)
#if scorex_vsone is None:
# TODO: add spatial verification as a filter score
# augment the vsone scores
# TODO: paramaterize
weighted_ave_score = True
if weighted_ave_score:
# weighted average scoring
new_fs_vsone = vt.weighted_average_scoring(new_fsv_vsone,
weight_filtxs,
nonweight_filtxs)
else:
# product scoring
new_fs_vsone = product_scoring(new_fsv_vsone)
new_score_vsone = new_fs_vsone.sum()
qres_vsone.aid2_fsv[daid] = new_fsv_vsone
qres_vsone.aid2_fs[daid] = new_fs_vsone
qres_vsone.aid2_score[daid] = new_score_vsone
# FIXME: this is not how to compute new probability
#if qres_vsone.aid2_prob is not None:
# qres_vsone.aid2_prob[daid] = qres_vsone.aid2_score[daid]
# This is how to compute new probability
if qreq_vsone_.qparams.score_normalization:
# FIXME: TODO: Have unsupported scores be represented as Nones
# while score normalizer is still being trained.
normalizer = qreq_vsone_.normalizer
daid2_score = qres_vsone.aid2_score
score_list = list(six.itervalues(daid2_score))
daid_list = list(six.iterkeys(daid2_score))
prob_list = normalizer.normalize_score_list(score_list)
daid2_prob = dict(zip(daid_list, prob_list))
qres_vsone.aid2_prob = daid2_prob
def apply_new_qres_filter_scores(qreq_vsone_, qres_vsone, newfsv_list, newscore_aids, filtkey):
r"""
applies the new filter scores vectors to a query result and updates other
scores
Args:
qres_vsone (QueryResult): object of feature correspondences and scores
newfsv_list (list):
newscore_aids (?):
filtkey (?):
CommandLine:
python -m ibeis.algo.hots.special_query --test-apply_new_qres_filter_scores
Example:
>>> # DISABLE_DOCTEST
>>> from ibeis.algo.hots.special_query import * # NOQA
>>> ibs, valid_aids = testdata_special_query()
>>> qaids = valid_aids[0:1]
>>> daids = valid_aids[1:]
>>> qaid = qaids[0]
>>> filtkey = hstypes.FiltKeys.DISTINCTIVENESS
>>> use_cache = False
>>> qaid2_qres_vsmany, qreq_vsmany_ = query_vsmany_initial(ibs, qaids, daids, use_cache)
>>> vsone_query_pairs = build_vsone_shortlist(ibs, qaid2_qres_vsmany)
>>> qaid2_qres_vsone, qreq_vsone_ = query_vsone_pairs(ibs, vsone_query_pairs, use_cache)
>>> qreq_vsone_.load_score_normalizer()
>>> qres_vsone = qaid2_qres_vsone[qaid]
>>> qres_vsmany = qaid2_qres_vsmany[qaid]
>>> top_aids = vsone_query_pairs[0][1]
>>> newfsv_list, newscore_aids = get_new_qres_distinctiveness(qres_vsone, qres_vsmany, top_aids, filtkey)
>>> apply_new_qres_filter_scores(qreq_vsone_, qres_vsone, newfsv_list, newscore_aids, filtkey)
Ignore:
qres_vsone.show_top(ibs, name_scoring=True)
print(qres_vsone.get_inspect_str(ibs=ibs, name_scoring=True))
print(qres_vsmany.get_inspect_str(ibs=ibs, name_scoring=True))
"""
assert ut.listfind(qres_vsone.filtkey_list, filtkey) is None
# HACK to update result cfgstr
qres_vsone.filtkey_list.append(filtkey)
qres_vsone.cfgstr = qreq_vsone_.get_cfgstr()
# Find positions of weight filters and score filters
# so we can apply a weighted average
#numer_filters = [hstypes.FiltKeys.LNBNN, hstypes.FiltKeys.RATIO]
weight_filters = hstypes.WEIGHT_FILTERS
weight_filtxs, nonweight_filtxs = vt.index_partition(qres_vsone.filtkey_list, weight_filters)
for new_fsv_vsone, daid in zip(newfsv_list, newscore_aids):
#scorex_vsone = ut.listfind(qres_vsone.filtkey_list, filtkey)
#if scorex_vsone is None:
# TODO: add spatial verification as a filter score
# augment the vsone scores
# TODO: paramaterize
weighted_ave_score = True
if weighted_ave_score:
# weighted average scoring
new_fs_vsone = vt.weighted_average_scoring(new_fsv_vsone, weight_filtxs, nonweight_filtxs)
else:
# product scoring
new_fs_vsone = product_scoring(new_fsv_vsone)
new_score_vsone = new_fs_vsone.sum()
qres_vsone.aid2_fsv[daid] = new_fsv_vsone
qres_vsone.aid2_fs[daid] = new_fs_vsone
qres_vsone.aid2_score[daid] = new_score_vsone
# FIXME: this is not how to compute new probability
#if qres_vsone.aid2_prob is not None:
# qres_vsone.aid2_prob[daid] = qres_vsone.aid2_score[daid]
# This is how to compute new probability
if qreq_vsone_.qparams.score_normalization:
# FIXME: TODO: Have unsupported scores be represented as Nones
# while score normalizer is still being trained.
normalizer = qreq_vsone_.normalizer
daid2_score = qres_vsone.aid2_score
score_list = list(six.itervalues(daid2_score))
daid_list = list(six.iterkeys(daid2_score))
prob_list = normalizer.normalize_score_list(score_list)
daid2_prob = dict(zip(daid_list, prob_list))
qres_vsone.aid2_prob = daid2_prob
