def learn_visual_words(annots_df, taids, nWords, use_cache=USE_CACHE_WORDS,
with_pandas=WITH_PANDAS):
"""
Computes visual words
>>> from ibeis.model.hots.smk.smk_index import * # NOQA
>>> from ibeis.model.hots.smk import smk_debug
>>> ibs, annots_df, taids, daids, qaids, nWords = smk_debug.testdata_dataframe()
>>> use_cache = USE_CACHE_WORDS
>>> words = learn_visual_words(annots_df, taids, nWords)
>>> print(words.shape)
(8000, 128)
"""
max_iters = 200
flann_params = {}
train_vecs_list = [pdh.ensure_values(vecs) for vecs in annots_df['vecs'][taids].values]
train_vecs = np.vstack(train_vecs_list)
print('Training %d word vocabulary with %d annots and %d descriptors' %
(nWords, len(taids), len(train_vecs)))
kwds = dict(max_iters=max_iters, use_cache=use_cache, appname='smk',
flann_params=flann_params)
_words = clustertool.cached_akmeans(train_vecs, nWords, **kwds)
if with_pandas:
# Pandasify
wx_series = pdh.RangeIndex(len(_words), name='wx')
#words = pd.DataFrame(_words, index=wx_series, columns=VEC_COLUMNS)
words = pd.DataFrame(_words, index=wx_series)
else:
words = _words
return words
def learn_visual_words(annots_df, train_aids, nCentroids):
vecs_list = annots_df['vecs'][train_aids].as_matrix()
train_vecs = np.vstack(vecs_list)
print('Training %d word vocabulary with %d annots and %d descriptors' %
(nCentroids, len(train_aids), len(train_vecs)))
words = clustertool.cached_akmeans(train_vecs, nCentroids, max_iters=100)
return words
def train_paris_vocab(ibs):
"""
CommandLine:
python dev.py --db Paris --cmd
"""
# UNFINISHED
aid_list = []
# use only one annotion per image
for aids in ibs.get_image_aids(ibs.get_valid_gids()):
if len(aids) == 1:
aid_list.append(aids[0])
else:
# use annote with largest area
aid_list.append(aids[np.argmax(ibs.get_annot_bbox_area(aids))])
vecs_list = ibs.get_annot_vecs(aid_list)
vecs = np.vstack(vecs_list)
nWords = 8000
from vtool import clustering2 as clustertool
print('vecs are: %r' % utool.get_object_size_str(vecs))
_words = clustertool.cached_akmeans(vecs, nWords, max_iters=500, use_cache=True, appname='smk') # NOQA
vec_mean = vecs.mean(axis=0).astype(np.float32)
vec_mean.shape = (1, vec_mean.shape[0])
vecs_centered = vecs - vec_mean
norm_ = npl.norm(arr1, axis=1)
norm_.shape = (norm_.size, 1)
vecs_norm = np.divide(arr1, norm_) # , out=out)
print('vecs_centered are: %r' % utool.get_object_size_str(vecs_centered))
vecs_post = np.round(128 * np.sqrt(np.abs(vecs_norm)) * np.sign(vecs_norm)).astype(np.int8) # NOQA
def learn_visual_words(annots_df, taids, nWords, use_cache=USE_CACHE_WORDS):
"""
Computes visual words
>>> from ibeis.model.hots.smk.smk_index import * # NOQA
>>> from ibeis.model.hots.smk import smk_debug
>>> ibs, annots_df, taids, daids, qaids, nWords = smk_debug.testdata_dataframe()
>>> use_cache = USE_CACHE_WORDS
>>> words = learn_visual_words(annots_df, taids, nWords)
>>> print(words.shape)
(8000, 128)
"""
max_iters = 200
flann_params = {}
train_vecs_list = [
pdh.ensure_values(vecs) for vecs in annots_df['vecs'][taids].values
]
train_vecs = np.vstack(train_vecs_list)
print('Training %d word vocabulary with %d annots and %d descriptors' %
(nWords, len(taids), len(train_vecs)))
kwds = dict(max_iters=max_iters,
use_cache=use_cache,
appname='smk',
flann_params=flann_params)
_words = clustertool.cached_akmeans(train_vecs, nWords, **kwds)
if WITH_PANDAS:
# Pandasify
wx_series = pdh.RangeIndex(len(_words), name='wx')
#words = pd.DataFrame(_words, index=wx_series, columns=VEC_COLUMNS)
words = pd.DataFrame(_words, index=wx_series)
else:
words = _words
return words
def learn_visual_words(annots_df, taids, nWords, use_cache=True):
"""
Computes visual words
>>> from ibeis.model.hots.smk.smk import * # NOQA
>>> ibs, annots_df, taids, daids, qaids, nWords = testdata()
>>> words = learn_visual_words(annots_df, taids, nWords)
"""
vecs_list = annots_df['vecs'][taids].as_matrix()
train_vecs = np.vstack(vecs_list)
print('Training %d word vocabulary with %d annots and %d descriptors' %
(nWords, len(taids), len(train_vecs)))
cache_dir = utool.get_app_resource_dir('smk')
words = clustertool.cached_akmeans(train_vecs, nWords, max_iters=100,
use_cache=use_cache, cache_dir=cache_dir)
return words
def learn_visual_words(annots_df, taids, nWords, use_cache=True):
"""
Computes visual words
>>> from ibeis.model.hots.smk.smk import * # NOQA
>>> ibs, annots_df, taids, daids, qaids, nWords = testdata()
>>> words = learn_visual_words(annots_df, taids, nWords)
"""
vecs_list = annots_df['vecs'][taids].as_matrix()
train_vecs = np.vstack(vecs_list)
print('Training %d word vocabulary with %d annots and %d descriptors' %
(nWords, len(taids), len(train_vecs)))
cache_dir = utool.get_app_resource_dir('smk')
words = clustertool.cached_akmeans(train_vecs,
nWords,
max_iters=100,
use_cache=use_cache,
cache_dir=cache_dir)
return words
def learn_visual_words(ibs, config2_=None, use_cache=USE_CACHE_WORDS, memtrack=None):
"""
Computes and caches visual words
Args:
ibs (?):
qreq_ (QueryRequest): query request object with hyper-parameters
use_cache (bool): turns on disk based caching(default = True)
memtrack (None): (default = None)
Returns:
ndarray[uint8_t, ndim=2]: words - aggregate descriptor cluster centers
Returns:
words
Example:
>>> # SLOW_DOCTEST
>>> from ibeis.algo.hots.smk.smk_index import * # NOQA
>>> from ibeis.algo.hots.smk import smk_debug
>>> ibs, annots_df, taids, daids, qaids, qreq_, nWords = smk_debug.testdata_dataframe()
>>> use_cache = True
>>> words = learn_visual_words(ibs, qreq_)
>>> print(words.shape)
(8000, 128)
Example:
>>> # SLOW_DOCTEST
>>> from ibeis.algo.hots.smk.smk_index import * # NOQA
>>> import ibeis
>>> ibs = ibeis.opendb('PZ_Master1')
>>> config2_ = ibs.new_query_params(cfgdict=dict(nWords=128000))
>>> use_cache = True
>>> words = learn_visual_words(ibs, config2_)
>>> print(words.shape)
(8000, 128)
Auto:
from ibeis.algo.hots.smk import smk_index
import utool as ut
argdoc = ut.make_default_docstr(smk_index.learn_visual_words)
print(argdoc)
"""
#if memtrack is None:
# memtrack = ut.MemoryTracker('[learn_visual_words]')
#config2_ = qreq_.get_external_data_config2()
nWords = config2_.nWords
# TODO: Incorporated taids (vocab training ids) into qreq
if config2_.vocab_taids == 'all':
taids = ibs.get_valid_aids(species=ibs.get_primary_database_species()) # exemplar
else:
taids = config2_.vocab_taids
initmethod = config2_.vocab_init_method
max_iters = config2_.vocab_nIters
flann_params = config2_.vocab_flann_params
train_vecs_list = ibs.get_annot_vecs(taids, eager=True, config2_=config2_)
#memtrack.track_obj(train_vecs_list[0], 'train_vecs_list[0]')
#memtrack.report('loaded trainvecs')
train_vecs = np.vstack(train_vecs_list)
#memtrack.track_obj(train_vecs, 'train_vecs')
#memtrack.report('stacked trainvecs')
del train_vecs_list
print('[smk_index] Train Vocab(nWords=%d) using %d annots and %d descriptors' %
(nWords, len(taids), len(train_vecs)))
kwds = dict(max_iters=max_iters, use_cache=use_cache,
initmethod=initmethod, appname='smk',
flann_params=flann_params)
words = clustertool.cached_akmeans(train_vecs, nWords, **kwds)
del train_vecs
del kwds
#memtrack.report('returning words')
#del train_vecs_list
return words
def learn_visual_words(ibs, config2_=None, use_cache=USE_CACHE_WORDS, memtrack=None):
"""
Computes and caches visual words
Args:
ibs (?):
qreq_ (QueryRequest): query request object with hyper-parameters
use_cache (bool): turns on disk based caching(default = True)
memtrack (None): (default = None)
Returns:
ndarray[uint8_t, ndim=2]: words - aggregate descriptor cluster centers
Returns:
words
Example:
>>> # SLOW_DOCTEST
>>> from ibeis.algo.hots.smk.smk_index import * # NOQA
>>> from ibeis.algo.hots.smk import smk_debug
>>> ibs, annots_df, taids, daids, qaids, qreq_, nWords = smk_debug.testdata_dataframe()
>>> use_cache = True
>>> words = learn_visual_words(ibs, qreq_)
>>> print(words.shape)
(8000, 128)
Example:
>>> # SLOW_DOCTEST
>>> from ibeis.algo.hots.smk.smk_index import * # NOQA
>>> import ibeis
>>> ibs = ibeis.opendb('PZ_Master1')
>>> config2_ = ibs.new_query_params(cfgdict=dict(nWords=128000))
>>> use_cache = True
>>> words = learn_visual_words(ibs, config2_)
>>> print(words.shape)
(8000, 128)
Auto:
from ibeis.algo.hots.smk import smk_index
import utool as ut
argdoc = ut.make_default_docstr(smk_index.learn_visual_words)
print(argdoc)
"""
#if memtrack is None:
# memtrack = ut.MemoryTracker('[learn_visual_words]')
#config2_ = qreq_.extern_data_config2
nWords = config2_.nWords
# TODO: Incorporated taids (vocab training ids) into qreq
if config2_.vocab_taids == 'all':
taids = ibs.get_valid_aids(species=ibs.get_primary_database_species()) # exemplar
else:
taids = config2_.vocab_taids
initmethod = config2_.vocab_init_method
max_iters = config2_.vocab_nIters
flann_params = config2_.vocab_flann_params
train_vecs_list = ibs.get_annot_vecs(taids, eager=True, config2_=config2_)
#memtrack.track_obj(train_vecs_list[0], 'train_vecs_list[0]')
#memtrack.report('loaded trainvecs')
train_vecs = np.vstack(train_vecs_list)
#memtrack.track_obj(train_vecs, 'train_vecs')
#memtrack.report('stacked trainvecs')
del train_vecs_list
print('[smk_index] Train Vocab(nWords=%d) using %d annots and %d descriptors' %
(nWords, len(taids), len(train_vecs)))
kwds = dict(max_iters=max_iters, use_cache=use_cache,
initmethod=initmethod, appname='smk',
flann_params=flann_params)
words = clustertool.cached_akmeans(train_vecs, nWords, **kwds)
del train_vecs
del kwds
#memtrack.report('returning words')
#del train_vecs_list
return words