def main(args):
query_files = sorted(glob.glob(os.path.join(args.queries_path, '*.jpg')))
image_files = sorted(glob.glob(os.path.join(args.images_path, '*.jpg')))
if args.mode == 'eval':
with open(args.corresp_file, 'rb') as f:
query_gt = dict(pickle.load(f))
keypoint_methods = ['sift']
descriptor_methods = ['sift']
match_methods = ['brute_force']
distance_metrics = ['l1']
for keypoint_method, descriptor_method, match_method, distance_metric in product(
keypoint_methods, descriptor_methods, match_methods,
distance_metrics):
results = query_batch(query_files, image_files, keypoint_method,
descriptor_method, match_method, distance_metric)
if args.mode == 'eval':
actual = []
predicted = []
for query_file, result in zip(query_files, results):
actual.append(query_gt[_filename_to_id(query_file)])
image_files, scores = zip(*result)
if scores[0] == 0:
predicted.append([-1])
else:
predicted.append([
_filename_to_id(image_file)
for image_file in image_files
])
print('{}: {}'.format(_filename_to_id(query_file),
[(_filename_to_id(image_file), score)
for image_file, score in result][:5]))
print('MAP@{}: {}'.format(10, mapk(actual, predicted, 10)))
print('MAP@{}: {}'.format(5, mapk(actual, predicted, 5)))
print('MAP@{}: {}'.format(3, mapk(actual, predicted, 3)))
print('MAP@{}: {}'.format(1, mapk(actual, predicted, 1)))
elif args.mode == 'test':
predicted = []
for query_file, result in zip(query_files, results):
image_files, scores = zip(*result)
if scores[0] == 0:
predicted.append([-1])
else:
predicted.append([
_filename_to_id(image_file)
for image_file in image_files
])
_save_results(predicted,
args.results_path,
method='{}_{}_{}'.format(keypoint_method,
descriptor_method,
distance_metric))
else:
raise ValueError('Invalid mode.')
def test_mapk():
# Single entry
np.testing.assert_almost_equal([mapk([1], [[2, 3, 1, 1]], 'full')[0]],
[0.41666666])
# Multiple entries
np.testing.assert_almost_equal(
[mapk([1, 1], [[2, 3, 1, 1], [1, 1]], 'full')[0]], [0.70833333])
def main(args):
query_files = sorted(glob.glob(args.queries_path))
image_files = sorted(glob.glob(args.images_path))
if args.mode == 'eval':
with open(args.corresp_file, 'rb') as f:
query_gt = pickle.load(f)
color_methods = [
'rgb_histogram', 'hsv_histogram', 'lab_histogram', 'ycrcb_histogram',
'cld', 'rgb_histogram_pyramid', 'hsv_histogram_pyramid',
'lab_histogram_pyramid', 'ycrcb_histogram_pyramid'
]
texture_methods = [None, 'gabor', 'glcm']
dist_metrics = ['euclidean_distance', 'l1_distance', 'cosine_distance']
hist_metrics = [
'intersection', 'correlation', 'chi_square', 'hellinguer_distance',
'bhattacharya_distance'
]
for texture_method, color_method, metric in product(
texture_methods, color_methods, dist_metrics + hist_metrics):
print('({}, {}, {})'.format(color_method, texture_method, metric))
with Timer('query_batch'):
results = query_batch(query_files, image_files, color_method,
metric, texture_method)
if args.mode == 'eval':
actual = []
predicted = []
for query_file, result in zip(query_files, results):
actual.append([query_gt[_filename_to_id(query_file)]])
predicted.append([
_filename_to_id(image_file) for image_file, dist in result
])
print('MAP@K: {}'.format(mapk(actual, predicted)))
elif args.mode == 'test':
predicted = []
for query_file, result in zip(query_files, results):
predicted.append([
_filename_to_id(image_file) for image_file, dist in result
])
_save_results(predicted,
args.results_path,
method='{}_{}_{}'.format(color_method,
texture_method, metric))
else:
raise ValueError('Invalid mode.')
text_list = None
if args.text:
with open(os.path.join(query_path, "text_boxes.pkl"), 'rb') as file:
text_list = pickle.load(file)
result_list_of_lists = search_batch(query_list=query_list,
mask_list=mask_list,
descriptor=args.descriptor,
metric=args.metric,
bins=args.bins,
k=k,
multiple=args.multiple,
text_list=text_list)
map_k = mapk(ground_truth, result_list_of_lists, k=k)
print(f'map@{k} for the current run is {map_k}')
if k > 1:
map_1 = mapk(ground_truth, result_list_of_lists, k=1)
print(f'map@{1} for the current run is {map_1}')
# extra functions for pickling and visualizing results
# use -p in cmdline args
if args.pickle:
print(f'writing this list to results.pkl \n {result_list_of_lists}')
with open('result.pkl', 'wb') as f:
pickle.dump(result_list_of_lists, f)
# use -v in cmdline args
if args.plot:
query_index = np.random.randint(0, 30)
