def similar_images(self, img, algo):
if algo == 'ConvNet':
features = img.features('convnet')
db_features = pickle.load(open('../../datasets/Mnist/features/convnet/train_features.p', 'rb'))
# id_best_features = euclidean_distances(features, db_features)
id_best_features = oasis_similarity('mnist', 'convnet', features, db_features)
features_names_and_labels = pickle.load(open('../../datasets/Mnist/features/convnet/features_names_and_labels.p', 'rb'))
names = features_names_and_labels['names']
best_imgs = np.array(names)[id_best_features]
# images = self.convnet_oasis_similarity(features, db_features, names)
elif algo == 'BOW':
features = img.features('bag', 64)
samples = np.array(pickle.load(open('../../datasets/Mnist/features/bag/train_features.p', 'rb')))
db_features = []
names = []
for sample in samples:
db_features.append(sample['features'])
# names.append(sample['label'])
names.append(sample['name'])
db_features = np.array(db_features)
# id_best_features = self.euclidean_distances(features, db_features)
id_best_features = oasis_similarity('mnist', 'bag', features, db_features)
# images = convnet_oasis_similarity(features, db_features, names)
best_imgs = np.array(names)[id_best_features]
images = []
for img in best_imgs:
print(img)
img = img.replace('jpeg', 'jpg')
img_path = {
'path': settings.STATIC_URL + 'images/Mnist/' + img
}
images.append(img_path)
return images
def precision(dataset, algo, distance):
if dataset == 'mnist':
if algo == 'convnet':
db_features = pickle.load(open('../../datasets/Mnist/features/convnet/train_features.p', 'rb'))
labels = pickle.load(open('../../datasets/Mnist/features/convnet/features_names_and_labels.p', 'rb'))['labels']
if distance == 'oasis':
all_filenames = glob.glob('../../datasets/Mnist/features/convnet/test_features/*')
random.shuffle(all_filenames)
filenames = all_filenames[0:3]
precisions = []
recalls = []
for filename in filenames:
label = int(filename.split('/')[-1].split('_')[0])
nb_label = (label==labels).sum()
query_features = pickle.load(open(filename, 'rb'))
query_precisions = []
query_recalls = []
for max_rank in range(1,1500):
id_best_features = oasis_similarity(dataset, algo, query_features, db_features, max_rank)
best_labels = np.array(labels)[id_best_features]
precision = (best_labels == label).sum()/len(best_labels)
recall = (best_labels == label).sum()/nb_label
query_precisions.append(precision)
query_recalls.append(recall)
precisions.append(query_precisions)
recalls.append(query_recalls)
mean_precisions = np.mean(precisions, axis=1)
pickle.dump(mean_precisions, open('convnet_oasis_mean_precisions.p', 'wb'))
pickle.dump(recalls, open('convnet_oasis_recalls_at_ranks.p', 'wb'))
pickle.dump(precisions, open('convnet_oasis_precisions_at_ranks.p', 'wb'))
elif algo == 'bag':
samples = np.array(pickle.load(open('../../datasets/Mnist/features/bag/train_features.p', 'rb')))
db_features = []
labels = []
for sample in samples:
db_features.append(sample['features'])
labels.append(sample['label'])
# names.append(sample['name'])
db_features = np.array(db_features)
if distance == 'oasis':
all_filenames = glob.glob('../../datasets/Mnist/images/test/*')
random.shuffle(all_filenames)
filenames = all_filenames[0:3]
f = gzip.open('../../datasets/Mnist/mnist.pkl.gz', 'rb')
train_set, valid_set, test_set = pickle.load(f)
f.close()
X_test, y_test = test_set
combined = list(zip(X_test, y_test))
random.shuffle(combined)
X_test[:], y_test[:] = zip(*combined)
X_test = X_test[0:3]
y_test = y_test[0:3]
precisions = []
recalls = []
for img, label in zip(X_test, y_test):
print(label)
nb_label = (label==labels).sum()
print(nb_label)
# img = imread(filename)
query_features = bag_of_words(img, 64)
# label = int(filename.split('/')[-1].split('_')[0])
query_precisions = []
query_recalls = []
for max_rank in range(1,1500):
id_best_features = oasis_similarity(dataset, algo, query_features, db_features, max_rank)
best_labels = np.array(labels)[id_best_features]
precision = (best_labels == label).sum()/len(best_labels)
recall = (best_labels == label).sum()/nb_label
query_precisions.append(precision)
query_recalls.append(recall)
precisions.append(query_precisions)
recalls.append(query_recalls)
mean_precisions = np.mean(precisions, axis=1)
pickle.dump(mean_precisions, open('bag_oasis_mean_precisions.p', 'wb'))
pickle.dump(recalls, open('bag_oasis_recalls_at_ranks.p', 'wb'))
pickle.dump(precisions, open('bag_oasis_precisions_at_ranks.p', 'wb'))
