first push
This commit is contained in:
0
cirtorch/utils/__init__.py
Executable file
0
cirtorch/utils/__init__.py
Executable file
154
cirtorch/utils/download.py
Executable file
154
cirtorch/utils/download.py
Executable file
@ -0,0 +1,154 @@
|
||||
import os
|
||||
|
||||
def download_test(data_dir):
|
||||
"""
|
||||
DOWNLOAD_TEST Checks, and, if required, downloads the necessary datasets for the testing.
|
||||
|
||||
download_test(DATA_ROOT) checks if the data necessary for running the example script exist.
|
||||
If not it downloads it in the folder structure:
|
||||
DATA_ROOT/test/oxford5k/ : folder with Oxford images and ground truth file
|
||||
DATA_ROOT/test/paris6k/ : folder with Paris images and ground truth file
|
||||
DATA_ROOT/test/roxford5k/ : folder with Oxford images and revisited ground truth file
|
||||
DATA_ROOT/test/rparis6k/ : folder with Paris images and revisited ground truth file
|
||||
"""
|
||||
|
||||
# Create data folder if it does not exist
|
||||
if not os.path.isdir(data_dir):
|
||||
os.mkdir(data_dir)
|
||||
|
||||
# Create datasets folder if it does not exist
|
||||
datasets_dir = os.path.join(data_dir, 'test')
|
||||
print('***************', os.path.exists(datasets_dir))
|
||||
#print(not os.path.isdir(datasets_dir))
|
||||
if not os.path.exists(datasets_dir):
|
||||
os.mkdir(datasets_dir)
|
||||
|
||||
# Download datasets folders test/DATASETNAME/
|
||||
datasets = ['oxford5k', 'paris6k', 'roxford5k', 'rparis6k']
|
||||
for di in range(len(datasets)):
|
||||
dataset = datasets[di]
|
||||
|
||||
if dataset == 'oxford5k':
|
||||
src_dir = 'http://www.robots.ox.ac.uk/~vgg/data/oxbuildings'
|
||||
dl_files = ['oxbuild_images.tgz']
|
||||
elif dataset == 'paris6k':
|
||||
src_dir = 'http://www.robots.ox.ac.uk/~vgg/data/parisbuildings'
|
||||
dl_files = ['paris_1.tgz', 'paris_2.tgz']
|
||||
elif dataset == 'roxford5k':
|
||||
src_dir = 'http://www.robots.ox.ac.uk/~vgg/data/oxbuildings'
|
||||
dl_files = ['oxbuild_images.tgz']
|
||||
elif dataset == 'rparis6k':
|
||||
src_dir = 'http://www.robots.ox.ac.uk/~vgg/data/parisbuildings'
|
||||
dl_files = ['paris_1.tgz', 'paris_2.tgz']
|
||||
else:
|
||||
raise ValueError('Unknown dataset: {}!'.format(dataset))
|
||||
|
||||
dst_dir = os.path.join(datasets_dir, dataset, 'jpg')
|
||||
print('%%%%%%%%%%%%%%%%',dst_dir, dataset)
|
||||
if not os.path.exists(dst_dir):
|
||||
# for oxford and paris download images
|
||||
if dataset == 'oxford5k' or dataset == 'paris6k':
|
||||
print('>> Dataset {} directory does not exist. Creating: {}'.format(dataset, dst_dir))
|
||||
os.makedirs(dst_dir)
|
||||
for dli in range(len(dl_files)):
|
||||
dl_file = dl_files[dli]
|
||||
src_file = os.path.join(src_dir, dl_file)
|
||||
dst_file = os.path.join(dst_dir, dl_file)
|
||||
print('>> Downloading dataset {} archive {}...'.format(dataset, dl_file))
|
||||
os.system('wget {} -O {}'.format(src_file, dst_file))
|
||||
print('>> Extracting dataset {} archive {}...'.format(dataset, dl_file))
|
||||
# create tmp folder
|
||||
dst_dir_tmp = os.path.join(dst_dir, 'tmp')
|
||||
os.system('mkdir {}'.format(dst_dir_tmp))
|
||||
# extract in tmp folder
|
||||
os.system('tar -zxf {} -C {}'.format(dst_file, dst_dir_tmp))
|
||||
# remove all (possible) subfolders by moving only files in dst_dir
|
||||
os.system('find {} -type f -exec mv -i {{}} {} \\;'.format(dst_dir_tmp, dst_dir))
|
||||
# remove tmp folder
|
||||
os.system('rm -rf {}'.format(dst_dir_tmp))
|
||||
print('>> Extracted, deleting dataset {} archive {}...'.format(dataset, dl_file))
|
||||
os.system('rm {}'.format(dst_file))
|
||||
|
||||
# for roxford and rparis just make sym links
|
||||
elif dataset == 'roxford5k' or dataset == 'rparis6k':
|
||||
print('>> Dataset {} directory does not exist. Creating: {}'.format(dataset, dst_dir))
|
||||
dataset_old = dataset[1:]
|
||||
dst_dir_old = os.path.join(datasets_dir, dataset_old, 'jpg')
|
||||
os.mkdir(os.path.join(datasets_dir, dataset))
|
||||
os.system('ln -s {} {}'.format(dst_dir_old, dst_dir))
|
||||
print('>> Created symbolic link from {} jpg to {} jpg'.format(dataset_old, dataset))
|
||||
|
||||
|
||||
gnd_src_dir = os.path.join('http://cmp.felk.cvut.cz/cnnimageretrieval/data', 'test', dataset)
|
||||
gnd_dst_dir = os.path.join(datasets_dir, dataset)
|
||||
gnd_dl_file = 'gnd_{}.pkl'.format(dataset)
|
||||
gnd_src_file = os.path.join(gnd_src_dir, gnd_dl_file)
|
||||
gnd_dst_file = os.path.join(gnd_dst_dir, gnd_dl_file)
|
||||
if not os.path.exists(gnd_dst_file):
|
||||
print('>> Downloading dataset {} ground truth file...'.format(dataset))
|
||||
os.system('wget {} -O {}'.format(gnd_src_file, gnd_dst_file))
|
||||
|
||||
|
||||
def download_train(data_dir):
|
||||
"""
|
||||
DOWNLOAD_TRAIN Checks, and, if required, downloads the necessary datasets for the training.
|
||||
|
||||
download_train(DATA_ROOT) checks if the data necessary for running the example script exist.
|
||||
If not it downloads it in the folder structure:
|
||||
DATA_ROOT/train/retrieval-SfM-120k/ : folder with rsfm120k images and db files
|
||||
DATA_ROOT/train/retrieval-SfM-30k/ : folder with rsfm30k images and db files
|
||||
"""
|
||||
|
||||
# Create data folder if it does not exist
|
||||
if not os.path.isdir(data_dir):
|
||||
os.mkdir(data_dir)
|
||||
|
||||
# Create datasets folder if it does not exist
|
||||
datasets_dir = os.path.join(data_dir, 'train')
|
||||
if not os.path.isdir(datasets_dir):
|
||||
os.mkdir(datasets_dir)
|
||||
|
||||
# Download folder train/retrieval-SfM-120k/
|
||||
src_dir = os.path.join('http://cmp.felk.cvut.cz/cnnimageretrieval/data', 'train', 'ims')
|
||||
dst_dir = os.path.join(datasets_dir, 'retrieval-SfM-120k', 'ims')
|
||||
dl_file = 'ims.tar.gz'
|
||||
if not os.path.isdir(dst_dir):
|
||||
src_file = os.path.join(src_dir, dl_file)
|
||||
dst_file = os.path.join(dst_dir, dl_file)
|
||||
print('>> Image directory does not exist. Creating: {}'.format(dst_dir))
|
||||
os.makedirs(dst_dir)
|
||||
print('>> Downloading ims.tar.gz...')
|
||||
os.system('wget {} -O {}'.format(src_file, dst_file))
|
||||
print('>> Extracting {}...'.format(dst_file))
|
||||
os.system('tar -zxf {} -C {}'.format(dst_file, dst_dir))
|
||||
print('>> Extracted, deleting {}...'.format(dst_file))
|
||||
os.system('rm {}'.format(dst_file))
|
||||
|
||||
# Create symlink for train/retrieval-SfM-30k/
|
||||
dst_dir_old = os.path.join(datasets_dir, 'retrieval-SfM-120k', 'ims')
|
||||
dst_dir = os.path.join(datasets_dir, 'retrieval-SfM-30k', 'ims')
|
||||
if not os.path.exists(dst_dir):
|
||||
os.makedirs(os.path.join(datasets_dir, 'retrieval-SfM-30k'))
|
||||
os.system('ln -s {} {}'.format(dst_dir_old, dst_dir))
|
||||
print('>> Created symbolic link from retrieval-SfM-120k/ims to retrieval-SfM-30k/ims')
|
||||
|
||||
# Download db files
|
||||
src_dir = os.path.join('http://cmp.felk.cvut.cz/cnnimageretrieval/data', 'train', 'dbs')
|
||||
datasets = ['retrieval-SfM-120k', 'retrieval-SfM-30k']
|
||||
for dataset in datasets:
|
||||
dst_dir = os.path.join(datasets_dir, dataset)
|
||||
if dataset == 'retrieval-SfM-120k':
|
||||
dl_files = ['{}.pkl'.format(dataset), '{}-whiten.pkl'.format(dataset)]
|
||||
elif dataset == 'retrieval-SfM-30k':
|
||||
dl_files = ['{}-whiten.pkl'.format(dataset)]
|
||||
|
||||
if not os.path.isdir(dst_dir):
|
||||
print('>> Dataset directory does not exist. Creating: {}'.format(dst_dir))
|
||||
os.mkdir(dst_dir)
|
||||
|
||||
for i in range(len(dl_files)):
|
||||
src_file = os.path.join(src_dir, dl_files[i])
|
||||
dst_file = os.path.join(dst_dir, dl_files[i])
|
||||
if not os.path.isfile(dst_file):
|
||||
print('>> DB file {} does not exist. Downloading...'.format(dl_files[i]))
|
||||
os.system('wget {} -O {}'.format(src_file, dst_file))
|
||||
152
cirtorch/utils/download_win.py
Executable file
152
cirtorch/utils/download_win.py
Executable file
@ -0,0 +1,152 @@
|
||||
import os
|
||||
|
||||
def download_test(data_dir):
|
||||
"""
|
||||
DOWNLOAD_TEST Checks, and, if required, downloads the necessary datasets for the testing.
|
||||
|
||||
download_test(DATA_ROOT) checks if the data necessary for running the example script exist.
|
||||
If not it downloads it in the folder structure:
|
||||
DATA_ROOT/test/oxford5k/ : folder with Oxford images and ground truth file
|
||||
DATA_ROOT/test/paris6k/ : folder with Paris images and ground truth file
|
||||
DATA_ROOT/test/roxford5k/ : folder with Oxford images and revisited ground truth file
|
||||
DATA_ROOT/test/rparis6k/ : folder with Paris images and revisited ground truth file
|
||||
"""
|
||||
|
||||
# Create data folder if it does not exist
|
||||
if not os.path.isdir(data_dir):
|
||||
os.mkdir(data_dir)
|
||||
|
||||
# Create datasets folder if it does not exist
|
||||
datasets_dir = os.path.join(data_dir, 'test')
|
||||
if not os.path.isdir(datasets_dir):
|
||||
os.mkdir(datasets_dir)
|
||||
|
||||
# Download datasets folders test/DATASETNAME/
|
||||
datasets = ['oxford5k', 'paris6k', 'roxford5k', 'rparis6k']
|
||||
for di in range(len(datasets)):
|
||||
dataset = datasets[di]
|
||||
|
||||
if dataset == 'oxford5k':
|
||||
src_dir = 'http://www.robots.ox.ac.uk/~vgg/data/oxbuildings'
|
||||
dl_files = ['oxbuild_images.tgz']
|
||||
elif dataset == 'paris6k':
|
||||
src_dir = 'http://www.robots.ox.ac.uk/~vgg/data/parisbuildings'
|
||||
dl_files = ['paris_1.tgz', 'paris_2.tgz']
|
||||
elif dataset == 'roxford5k':
|
||||
src_dir = 'http://www.robots.ox.ac.uk/~vgg/data/oxbuildings'
|
||||
dl_files = ['oxbuild_images.tgz']
|
||||
elif dataset == 'rparis6k':
|
||||
src_dir = 'http://www.robots.ox.ac.uk/~vgg/data/parisbuildings'
|
||||
dl_files = ['paris_1.tgz', 'paris_2.tgz']
|
||||
else:
|
||||
raise ValueError('Unknown dataset: {}!'.format(dataset))
|
||||
|
||||
dst_dir = os.path.join(datasets_dir, dataset, 'jpg')
|
||||
if not os.path.isdir(dst_dir):
|
||||
|
||||
# for oxford and paris download images
|
||||
if dataset == 'oxford5k' or dataset == 'paris6k':
|
||||
print('>> Dataset {} directory does not exist. Creating: {}'.format(dataset, dst_dir))
|
||||
os.makedirs(dst_dir)
|
||||
for dli in range(len(dl_files)):
|
||||
dl_file = dl_files[dli]
|
||||
src_file = os.path.join(src_dir, dl_file)
|
||||
dst_file = os.path.join(dst_dir, dl_file)
|
||||
print('>> Downloading dataset {} archive {}...'.format(dataset, dl_file))
|
||||
os.system('wget {} -O {}'.format(src_file, dst_file))
|
||||
print('>> Extracting dataset {} archive {}...'.format(dataset, dl_file))
|
||||
# create tmp folder
|
||||
dst_dir_tmp = os.path.join(dst_dir, 'tmp')
|
||||
os.system('mkdir {}'.format(dst_dir_tmp))
|
||||
# extract in tmp folder
|
||||
os.system('tar -zxf {} -C {}'.format(dst_file, dst_dir_tmp))
|
||||
# remove all (possible) subfolders by moving only files in dst_dir
|
||||
os.system('find {} -type f -exec mv -i {{}} {} \\;'.format(dst_dir_tmp, dst_dir))
|
||||
# remove tmp folder
|
||||
os.system('rd {}'.format(dst_dir_tmp))
|
||||
print('>> Extracted, deleting dataset {} archive {}...'.format(dataset, dl_file))
|
||||
os.system('del {}'.format(dst_file))
|
||||
|
||||
# for roxford and rparis just make sym links
|
||||
elif dataset == 'roxford5k' or dataset == 'rparis6k':
|
||||
print('>> Dataset {} directory does not exist. Creating: {}'.format(dataset, dst_dir))
|
||||
dataset_old = dataset[1:]
|
||||
dst_dir_old = os.path.join(datasets_dir, dataset_old, 'jpg')
|
||||
os.mkdir(os.path.join(datasets_dir, dataset))
|
||||
os.system('cmd /c mklink /d {} {}'.format(dst_dir_old, dst_dir))
|
||||
print('>> Created symbolic link from {} jpg to {} jpg'.format(dataset_old, dataset))
|
||||
|
||||
|
||||
gnd_src_dir = os.path.join('http://cmp.felk.cvut.cz/cnnimageretrieval/data', 'test', dataset)
|
||||
gnd_dst_dir = os.path.join(datasets_dir, dataset)
|
||||
gnd_dl_file = 'gnd_{}.pkl'.format(dataset)
|
||||
gnd_src_file = os.path.join(gnd_src_dir, gnd_dl_file)
|
||||
gnd_dst_file = os.path.join(gnd_dst_dir, gnd_dl_file)
|
||||
if not os.path.exists(gnd_dst_file):
|
||||
print('>> Downloading dataset {} ground truth file...'.format(dataset))
|
||||
os.system('wget {} -O {}'.format(gnd_src_file, gnd_dst_file))
|
||||
|
||||
|
||||
def download_train(data_dir):
|
||||
"""
|
||||
DOWNLOAD_TRAIN Checks, and, if required, downloads the necessary datasets for the training.
|
||||
|
||||
download_train(DATA_ROOT) checks if the data necessary for running the example script exist.
|
||||
If not it downloads it in the folder structure:
|
||||
DATA_ROOT/train/retrieval-SfM-120k/ : folder with rsfm120k images and db files
|
||||
DATA_ROOT/train/retrieval-SfM-30k/ : folder with rsfm30k images and db files
|
||||
"""
|
||||
|
||||
# Create data folder if it does not exist
|
||||
if not os.path.isdir(data_dir):
|
||||
os.mkdir(data_dir)
|
||||
print(data_dir)
|
||||
# Create datasets folder if it does not exist
|
||||
datasets_dir = os.path.join(data_dir, 'train')
|
||||
if not os.path.isdir(datasets_dir):
|
||||
os.mkdir(datasets_dir)
|
||||
|
||||
# Download folder train/retrieval-SfM-120k/
|
||||
src_dir = os.path.join('http://cmp.felk.cvut.cz/cnnimageretrieval/data', 'train', 'ims')
|
||||
dst_dir = os.path.join(datasets_dir, 'retrieval-SfM-120k', 'ims')
|
||||
dl_file = 'ims.tar.gz'
|
||||
if not os.path.isdir(dst_dir):
|
||||
src_file = os.path.join(src_dir, dl_file)
|
||||
dst_file = os.path.join(dst_dir, dl_file)
|
||||
print('>> Image directory does not exist. Creating: {}'.format(dst_dir))
|
||||
os.makedirs(dst_dir)
|
||||
print('>> Downloading ims.tar.gz...')
|
||||
# os.system('wget {} -O {}'.format(src_file, dst_file))
|
||||
print('>> Extracting {}...'.format(dst_file))
|
||||
os.system('tar -zxf {} -C {}'.format(dst_file, dst_dir))
|
||||
print('>> Extracted, deleting {}...'.format(dst_file))
|
||||
os.system('del {}'.format(dst_file))
|
||||
|
||||
# Create symlink for train/retrieval-SfM-30k/
|
||||
dst_dir_old = os.path.join(datasets_dir, 'retrieval-SfM-120k', 'ims')
|
||||
dst_dir = os.path.join(datasets_dir, 'retrieval-SfM-30k', 'ims')
|
||||
if not os.path.isdir(dst_dir):
|
||||
os.makedirs(os.path.join(datasets_dir, 'retrieval-SfM-30k','ims'))
|
||||
os.system('mklink {} {}'.format(dst_dir_old, dst_dir))
|
||||
print('>> Created symbolic link from retrieval-SfM-120k/ims to retrieval-SfM-30k/ims')
|
||||
|
||||
# Download db files
|
||||
src_dir = os.path.join('http://cmp.felk.cvut.cz/cnnimageretrieval/data', 'train', 'dbs')
|
||||
datasets = ['retrieval-SfM-120k', 'retrieval-SfM-30k']
|
||||
for dataset in datasets:
|
||||
dst_dir = os.path.join(datasets_dir, dataset)
|
||||
if dataset == 'retrieval-SfM-120k':
|
||||
dl_files = ['{}.pkl'.format(dataset), '{}-whiten.pkl'.format(dataset)]
|
||||
elif dataset == 'retrieval-SfM-30k':
|
||||
dl_files = ['{}-whiten.pkl'.format(dataset)]
|
||||
|
||||
if not os.path.isdir(dst_dir):
|
||||
print('>> Dataset directory does not exist. Creating: {}'.format(dst_dir))
|
||||
os.mkdir(dst_dir)
|
||||
|
||||
for i in range(len(dl_files)):
|
||||
src_file = os.path.join(src_dir, dl_files[i])
|
||||
dst_file = os.path.join(dst_dir, dl_files[i])
|
||||
if not os.path.isfile(dst_file):
|
||||
print('>> DB file {} does not exist. Downloading...'.format(dl_files[i]))
|
||||
os.system('wget {} -O {}'.format(src_file, dst_file))
|
||||
149
cirtorch/utils/evaluate.py
Executable file
149
cirtorch/utils/evaluate.py
Executable file
@ -0,0 +1,149 @@
|
||||
import numpy as np
|
||||
|
||||
def compute_ap(ranks, nres):
|
||||
"""
|
||||
Computes average precision for given ranked indexes.
|
||||
|
||||
Arguments
|
||||
---------
|
||||
ranks : zerro-based ranks of positive images
|
||||
nres : number of positive images
|
||||
|
||||
Returns
|
||||
-------
|
||||
ap : average precision
|
||||
"""
|
||||
|
||||
# number of images ranked by the system
|
||||
nimgranks = len(ranks)
|
||||
|
||||
# accumulate trapezoids in PR-plot
|
||||
ap = 0
|
||||
|
||||
recall_step = 1. / nres
|
||||
|
||||
for j in np.arange(nimgranks):
|
||||
rank = ranks[j]
|
||||
|
||||
if rank == 0:
|
||||
precision_0 = 1.
|
||||
else:
|
||||
precision_0 = float(j) / rank
|
||||
|
||||
precision_1 = float(j + 1) / (rank + 1)
|
||||
|
||||
ap += (precision_0 + precision_1) * recall_step / 2.
|
||||
|
||||
return ap
|
||||
|
||||
def compute_map(ranks, gnd, kappas=[]):
|
||||
"""
|
||||
Computes the mAP for a given set of returned results.
|
||||
|
||||
Usage:
|
||||
map = compute_map (ranks, gnd)
|
||||
computes mean average precsion (map) only
|
||||
|
||||
map, aps, pr, prs = compute_map (ranks, gnd, kappas)
|
||||
computes mean average precision (map), average precision (aps) for each query
|
||||
computes mean precision at kappas (pr), precision at kappas (prs) for each query
|
||||
|
||||
Notes:
|
||||
1) ranks starts from 0, ranks.shape = db_size X #queries
|
||||
2) The junk results (e.g., the query itself) should be declared in the gnd stuct array
|
||||
3) If there are no positive images for some query, that query is excluded from the evaluation
|
||||
"""
|
||||
|
||||
map = 0.
|
||||
nq = len(gnd) # number of queries
|
||||
aps = np.zeros(nq)
|
||||
pr = np.zeros(len(kappas))
|
||||
prs = np.zeros((nq, len(kappas)))
|
||||
nempty = 0
|
||||
|
||||
for i in np.arange(nq):
|
||||
qgnd = np.array(gnd[i]['ok'])
|
||||
|
||||
# no positive images, skip from the average
|
||||
if qgnd.shape[0] == 0:
|
||||
aps[i] = float('nan')
|
||||
prs[i, :] = float('nan')
|
||||
nempty += 1
|
||||
continue
|
||||
|
||||
try:
|
||||
qgndj = np.array(gnd[i]['junk'])
|
||||
except:
|
||||
qgndj = np.empty(0)
|
||||
|
||||
# sorted positions of positive and junk images (0 based)
|
||||
pos = np.arange(ranks.shape[0])[np.in1d(ranks[:,i], qgnd)]
|
||||
junk = np.arange(ranks.shape[0])[np.in1d(ranks[:,i], qgndj)]
|
||||
|
||||
k = 0;
|
||||
ij = 0;
|
||||
if len(junk):
|
||||
# decrease positions of positives based on the number of
|
||||
# junk images appearing before them
|
||||
ip = 0
|
||||
while (ip < len(pos)):
|
||||
while (ij < len(junk) and pos[ip] > junk[ij]):
|
||||
k += 1
|
||||
ij += 1
|
||||
pos[ip] = pos[ip] - k
|
||||
ip += 1
|
||||
|
||||
# compute ap
|
||||
ap = compute_ap(pos, len(qgnd))
|
||||
map = map + ap
|
||||
aps[i] = ap
|
||||
|
||||
# compute precision @ k
|
||||
pos += 1 # get it to 1-based
|
||||
for j in np.arange(len(kappas)):
|
||||
kq = min(max(pos), kappas[j]);
|
||||
prs[i, j] = (pos <= kq).sum() / kq
|
||||
pr = pr + prs[i, :]
|
||||
|
||||
map = map / (nq - nempty)
|
||||
pr = pr / (nq - nempty)
|
||||
|
||||
return map, aps, pr, prs
|
||||
|
||||
|
||||
def compute_map_and_print(dataset, ranks, gnd, kappas=[1, 5, 10]):
|
||||
|
||||
# old evaluation protocol
|
||||
if dataset.startswith('oxford5k') or dataset.startswith('paris6k'):
|
||||
map, aps, _, _ = compute_map(ranks, gnd)
|
||||
print('>> {}: mAP {:.2f}'.format(dataset, np.around(map*100, decimals=2)))
|
||||
|
||||
# new evaluation protocol
|
||||
elif dataset.startswith('roxford5k') or dataset.startswith('rparis6k'):
|
||||
|
||||
gnd_t = []
|
||||
for i in range(len(gnd)):
|
||||
g = {}
|
||||
g['ok'] = np.concatenate([gnd[i]['easy']])
|
||||
g['junk'] = np.concatenate([gnd[i]['junk'], gnd[i]['hard']])
|
||||
gnd_t.append(g)
|
||||
mapE, apsE, mprE, prsE = compute_map(ranks, gnd_t, kappas)
|
||||
|
||||
gnd_t = []
|
||||
for i in range(len(gnd)):
|
||||
g = {}
|
||||
g['ok'] = np.concatenate([gnd[i]['easy'], gnd[i]['hard']])
|
||||
g['junk'] = np.concatenate([gnd[i]['junk']])
|
||||
gnd_t.append(g)
|
||||
mapM, apsM, mprM, prsM = compute_map(ranks, gnd_t, kappas)
|
||||
|
||||
gnd_t = []
|
||||
for i in range(len(gnd)):
|
||||
g = {}
|
||||
g['ok'] = np.concatenate([gnd[i]['hard']])
|
||||
g['junk'] = np.concatenate([gnd[i]['junk'], gnd[i]['easy']])
|
||||
gnd_t.append(g)
|
||||
mapH, apsH, mprH, prsH = compute_map(ranks, gnd_t, kappas)
|
||||
|
||||
print('>> {}: mAP E: {}, M: {}, H: {}'.format(dataset, np.around(mapE*100, decimals=2), np.around(mapM*100, decimals=2), np.around(mapH*100, decimals=2)))
|
||||
print('>> {}: mP@k{} E: {}, M: {}, H: {}'.format(dataset, kappas, np.around(mprE*100, decimals=2), np.around(mprM*100, decimals=2), np.around(mprH*100, decimals=2)))
|
||||
34
cirtorch/utils/general.py
Executable file
34
cirtorch/utils/general.py
Executable file
@ -0,0 +1,34 @@
|
||||
import os
|
||||
import hashlib
|
||||
|
||||
def get_root():
|
||||
return os.path.join(os.path.dirname(os.path.dirname(os.path.dirname(os.path.realpath(__file__)))))
|
||||
|
||||
|
||||
def get_data_root():
|
||||
return os.path.join(get_root(), 'data')
|
||||
|
||||
|
||||
def htime(c):
|
||||
c = round(c)
|
||||
|
||||
days = c // 86400
|
||||
hours = c // 3600 % 24
|
||||
minutes = c // 60 % 60
|
||||
seconds = c % 60
|
||||
|
||||
if days > 0:
|
||||
return '{:d}d {:d}h {:d}m {:d}s'.format(days, hours, minutes, seconds)
|
||||
if hours > 0:
|
||||
return '{:d}h {:d}m {:d}s'.format(hours, minutes, seconds)
|
||||
if minutes > 0:
|
||||
return '{:d}m {:d}s'.format(minutes, seconds)
|
||||
return '{:d}s'.format(seconds)
|
||||
|
||||
|
||||
def sha256_hash(filename, block_size=65536, length=8):
|
||||
sha256 = hashlib.sha256()
|
||||
with open(filename, 'rb') as f:
|
||||
for block in iter(lambda: f.read(block_size), b''):
|
||||
sha256.update(block)
|
||||
return sha256.hexdigest()[:length-1]
|
||||
65
cirtorch/utils/whiten.py
Executable file
65
cirtorch/utils/whiten.py
Executable file
@ -0,0 +1,65 @@
|
||||
import os
|
||||
import numpy as np
|
||||
|
||||
def whitenapply(X, m, P, dimensions=None):
|
||||
|
||||
if not dimensions:
|
||||
dimensions = P.shape[0]
|
||||
|
||||
X = np.dot(P[:dimensions, :], X-m)
|
||||
X = X / (np.linalg.norm(X, ord=2, axis=0, keepdims=True) + 1e-6)
|
||||
|
||||
return X
|
||||
|
||||
def pcawhitenlearn(X):
|
||||
|
||||
N = X.shape[1]
|
||||
|
||||
# Learning PCA w/o annotations
|
||||
m = X.mean(axis=1, keepdims=True)
|
||||
Xc = X - m
|
||||
Xcov = np.dot(Xc, Xc.T)
|
||||
Xcov = (Xcov + Xcov.T) / (2*N)
|
||||
eigval, eigvec = np.linalg.eig(Xcov)
|
||||
order = eigval.argsort()[::-1]
|
||||
eigval = eigval[order]
|
||||
eigvec = eigvec[:, order]
|
||||
|
||||
P = np.dot(np.linalg.inv(np.sqrt(np.diag(eigval))), eigvec.T)
|
||||
|
||||
return m, P
|
||||
|
||||
def whitenlearn(X, qidxs, pidxs):
|
||||
|
||||
# Learning Lw w annotations
|
||||
m = X[:, qidxs].mean(axis=1, keepdims=True)
|
||||
df = X[:, qidxs] - X[:, pidxs]
|
||||
S = np.dot(df, df.T) / df.shape[1]
|
||||
P = np.linalg.inv(cholesky(S))
|
||||
df = np.dot(P, X-m)
|
||||
D = np.dot(df, df.T)
|
||||
eigval, eigvec = np.linalg.eig(D)
|
||||
order = eigval.argsort()[::-1]
|
||||
eigval = eigval[order]
|
||||
eigvec = eigvec[:, order]
|
||||
|
||||
P = np.dot(eigvec.T, P)
|
||||
|
||||
return m, P
|
||||
|
||||
def cholesky(S):
|
||||
# Cholesky decomposition
|
||||
# with adding a small value on the diagonal
|
||||
# until matrix is positive definite
|
||||
alpha = 0
|
||||
while 1:
|
||||
try:
|
||||
L = np.linalg.cholesky(S + alpha*np.eye(*S.shape))
|
||||
return L
|
||||
except:
|
||||
if alpha == 0:
|
||||
alpha = 1e-10
|
||||
else:
|
||||
alpha *= 10
|
||||
print(">>>> {}::cholesky: Matrix is not positive definite, adding {:.0e} on the diagonal"
|
||||
.format(os.path.basename(__file__), alpha))
|
||||
Reference in New Issue
Block a user