def _akmeans_iterate(data, centroids, datax2_clusterx_old, max_iters,
flann_params, ave_unchanged_thresh, ave_unchanged_iterwin):
""" Helper function which continues the iterations of akmeans """
num_data = data.shape[0]
num_clusters = centroids.shape[0]
# Keep track of how many points have changed in each iteration
xx2_unchanged = np.zeros(ave_unchanged_iterwin, dtype=centroids.dtype) + len(data)
print('[akmeans] Running akmeans: data.shape=%r ; num_clusters=%r' %
(data.shape, num_clusters))
print('[akmeans] * max_iters = %r ' % max_iters)
print('[akmeans] * ave_unchanged_iterwin=%r ; ave_unchanged_thresh=%r' %
(ave_unchanged_thresh, ave_unchanged_iterwin))
#print('[akmeans] Printing akmeans info in format: time (iterx, ave(#changed), #unchanged)')
xx = 0
for xx in range(0, max_iters):
tt = ut.tic()
ut.print_('...tic')
# 1) Find each datapoints nearest cluster center
(datax2_clusterx, _dist) = nn.ann_flann_once(centroids, data, 1, flann_params)
ellapsed = ut.toc(tt)
ut.print_('...toc(%.2fs)' % ellapsed)
# 2) Compute new cluster centers
centroids = _compute_cluster_centers(num_data, num_clusters, data, centroids, datax2_clusterx)
# 3) Check for convergence (no change of cluster index)
#ut.print_('+')
num_changed = (datax2_clusterx_old != datax2_clusterx).sum()
xx2_unchanged[xx % ave_unchanged_iterwin] = num_changed
ave_unchanged = xx2_unchanged.mean()
#ut.print_(' (%d, %.2f, %d)\n' % (xx, ave_unchanged, num_changed))
if ave_unchanged < ave_unchanged_thresh:
break
else: # Iterate
datax2_clusterx_old = datax2_clusterx
#if xx % 5 == 0:
# sys.stdout.flush()
if xx == max_iters:
print('[akmeans] * AKMEANS: converged in %d/%d iters' % (xx + 1, max_iters))
else:
print('[akmeans] * AKMEANS: reached the maximum iterations after in %d/%d iters' % (xx + 1, max_iters))
sys.stdout.flush()
return (datax2_clusterx, centroids)
def _compute_cluster_centers(num_data, num_clusters, data, centroids, datax2_clusterx):
""" Computes the cluster centers and stores output in the outvar: centroids.
This outvar is also returned """
# sort data by cluster
datax_sort = datax2_clusterx.argsort()
clusterx_sort = datax2_clusterx[datax_sort]
# group datapoints by cluster using a sliding grouping algorithm
_L = 0
clusterx2_dataLRx = ut.alloc_nones(num_clusters)
for _R in range(len(datax_sort) + 1): # Slide R
if _R == num_data or clusterx_sort[_L] != clusterx_sort[_R]:
clusterx2_dataLRx[clusterx_sort[_L]] = (_L, _R)
_L = _R
# Compute the centers of each group (cluster) of datapoints
ut.print_('+')
for clusterx, dataLRx in enumerate(clusterx2_dataLRx):
if dataLRx is None:
continue # ON EMPTY CLUSTER
(_L, _R) = dataLRx
# The cluster center is the mean of its datapoints
centroids[clusterx] = np.mean(data[datax_sort[_L:_R]], axis=0)
#centroids[clusterx] = np.array(np.round(centroids[clusterx]), dtype=np.uint8)
return centroids
def measure_feat_pairs(allres, orgtype='top_true'):
print('Measure ' + orgtype + ' pairs')
orgres = allres.__dict__[orgtype]
entropy_list = []
scale_list = []
score_list = []
lbl = 'Measuring ' + orgtype + ' pair '
fmt_str = utool.make_progress_fmt_str(len(orgres), lbl)
rank_skips = []
gt_skips = []
for ix, (qrid, rid, score, rank) in enumerate(orgres.iter()):
utool.print_(fmt_str % (ix + 1, ))
# Skip low ranks
if rank > 5:
rank_skips.append(qrid)
continue
other_rids = ibs.get_other_indexed_rids(qrid)
# Skip no groundtruth
if len(other_rids) == 0:
gt_skips.append(qrid)
continue
qres = qrid2_qres[qrid]
# Get matching feature indexes
fm = qres.cx2_fm[rid]
# Get their scores
fs = qres.cx2_fs[rid]
# Get matching descriptors
printDBG('\nfm.shape=%r' % (fm.shape, ))
desc1 = cx2_desc[qrid][fm[:, 0]]
desc2 = cx2_desc[rid][fm[:, 1]]
# Get matching keypoints
kpts1 = cx2_kpts[qrid][fm[:, 0]]
kpts2 = cx2_kpts[rid][fm[:, 1]]
# Get their scale
scale1_m = ktool.get_scales(kpts1)
scale2_m = ktool.get_scales(kpts2)
# Get their entropy
entropy1 = descriptor_entropy(desc1, bw_factor=1)
entropy2 = descriptor_entropy(desc2, bw_factor=1)
# Append to results
entropy_tup = np.array(zip(entropy1, entropy2))
scale_tup = np.array(zip(scale1_m, scale2_m))
entropy_tup = entropy_tup.reshape(len(entropy_tup), 2)
scale_tup = scale_tup.reshape(len(scale_tup), 2)
entropy_list.append(entropy_tup)
scale_list.append(scale_tup)
score_list.append(fs)
print('Skipped %d total.' % (len(rank_skips) + len(gt_skips), ))
print('Skipped %d for rank > 5, %d for no gt' % (
len(rank_skips),
len(gt_skips),
))
print(np.unique(map(len, entropy_list)))
def evstack(tup):
return np.vstack(tup) if len(tup) > 0 else np.empty((0, 2))
def ehstack(tup):
return np.hstack(tup) if len(tup) > 0 else np.empty((0, 2))
entropy_pairs = evstack(entropy_list)
scale_pairs = evstack(scale_list)
scores = ehstack(score_list)
print('\n * Measured %d pairs' % len(entropy_pairs))
return entropy_pairs, scale_pairs, scores
def measure_feat_pairs(allres, orgtype='top_true'):
print('Measure ' + orgtype + ' pairs')
orgres = allres.__dict__[orgtype]
entropy_list = []
scale_list = []
score_list = []
lbl = 'Measuring ' + orgtype + ' pair '
fmt_str = utool.make_progress_fmt_str(len(orgres), lbl)
rank_skips = []
gt_skips = []
for ix, (qrid, rid, score, rank) in enumerate(orgres.iter()):
utool.print_(fmt_str % (ix + 1,))
# Skip low ranks
if rank > 5:
rank_skips.append(qrid)
continue
other_rids = ibs.get_other_indexed_rids(qrid)
# Skip no groundtruth
if len(other_rids) == 0:
gt_skips.append(qrid)
continue
qres = qrid2_qres[qrid]
# Get matching feature indexes
fm = qres.cx2_fm[rid]
# Get their scores
fs = qres.cx2_fs[rid]
# Get matching descriptors
printDBG('\nfm.shape=%r' % (fm.shape,))
desc1 = cx2_desc[qrid][fm[:, 0]]
desc2 = cx2_desc[rid][fm[:, 1]]
# Get matching keypoints
kpts1 = cx2_kpts[qrid][fm[:, 0]]
kpts2 = cx2_kpts[rid][fm[:, 1]]
# Get their scale
scale1_m = ktool.get_scales(kpts1)
scale2_m = ktool.get_scales(kpts2)
# Get their entropy
entropy1 = descriptor_entropy(desc1, bw_factor=1)
entropy2 = descriptor_entropy(desc2, bw_factor=1)
# Append to results
entropy_tup = np.array(zip(entropy1, entropy2))
scale_tup = np.array(zip(scale1_m, scale2_m))
entropy_tup = entropy_tup.reshape(len(entropy_tup), 2)
scale_tup = scale_tup.reshape(len(scale_tup), 2)
entropy_list.append(entropy_tup)
scale_list.append(scale_tup)
score_list.append(fs)
print('Skipped %d total.' % (len(rank_skips) + len(gt_skips),))
print('Skipped %d for rank > 5, %d for no gt' % (len(rank_skips), len(gt_skips),))
print(np.unique(map(len, entropy_list)))
def evstack(tup):
return np.vstack(tup) if len(tup) > 0 else np.empty((0, 2))
def ehstack(tup):
return np.hstack(tup) if len(tup) > 0 else np.empty((0, 2))
entropy_pairs = evstack(entropy_list)
scale_pairs = evstack(scale_list)
scores = ehstack(score_list)
print('\n * Measured %d pairs' % len(entropy_pairs))
return entropy_pairs, scale_pairs, scores