def similar_memes(name):
ext = name.split('.')[1]
url = bottle.request.query['url']
print(url)
f = FEATURE(cv2.imdecode(np.frombuffer(requests.get(url).content, dtype=np.uint8), 1))
f = normalize_features(f.reshape((1, f.size)))
h = image_search.hash_samples(f, HASH_DATA['proj'], HASH_DATA['sample_mean'])[0]
similar = [(int(dist), IMAGE_URIS[ind], IMAGE_KYM_URIS[ind])
for dist, ind in HAMMING.knn(HASHES, h, 50)]
meme_counts = {}
for _, _, x in similar:
try:
meme_counts[x] += 1
except KeyError:
meme_counts[x] = 1
meme_counts = sorted(meme_counts.items(), key=lambda x: x[1], reverse=True)
out = {'similar': similar, 'meme_counts': meme_counts}
if ext == 'js':
return out
return ''.join('<img src="%s"/>' % x for _, x, _ in out['similar'])
def main():
image_uris, features = [], []
with open('features.pkl') as fp:
while 1:
try:
i, f = pickle.load(fp)
except:
break
image_uris.append(i)
features.append(f)
print(len(image_uris))
features = np.asfarray(features)
features = normalize_features(features)
train_samples, test_samples = image_search.samples_split_train_test(features)
out = image_search.evaluate_hashing_method(train_samples, test_samples, 128, 'RR')
print(image_search.print_pr_ret(out))
print(features.shape)
print(out['proj'].shape)
print(out['sample_mean'].shape)
hashes = image_search.hash_samples(features, out['proj'], out['sample_mean'])
pickle.dump((image_uris, hashes, out), open('hashes.pkl', 'w'), -1)
