def get_annotation(path, id):
db = leveldb.LevelDB(path)
datum_string = db.Get(id)
if not datum_string:
print "Not found {0}".format(id)
return None
return caffe.io.datum_to_array(Datum.FromString(datum_string))
def get_image(path, id):
with lmdb.open(path, readonly=True, lock=False) as env:
with env.begin() as txn:
datum_string = txn.get(id)
if not datum_string:
print "Not found {0}".format(id)
return None
return np.array(Image.open(StringIO(Datum.FromString(datum_string).data)),
dtype="uint8")
def mnist_lmdb_to_h5(src, tgt):
with lmdb.open(src, map_size=1099511627776, readonly=True) as lmdb_env:
numSamples = int(lmdb_env.stat()['entries'])
with lmdb_env.begin(write=False) as txn:
with txn.cursor() as cur:
with h5py.File(tgt, 'w') as fd:
fd.create_dataset('data', (numSamples, 1, 28, 28),
dtype=np.float32)
fd.create_dataset('label', (numSamples, ),
dtype=np.float32)
for key, val in cur:
d = Datum.FromString(val)
img = np.array(np.fromstring(
d.data, dtype=np.uint8).reshape(1, 28, 28),
dtype=np.float32) / 255.0
index = int(key)
fd['data'][index, :, :, :] = img
fd['label'][index] = float(d.label)
import cv2
from caffe.proto.caffe_pb2 import Datum
import numpy as np
import lmdb
from IPython import embed
env = lmdb.open('./mnist_test_lmdb')
with env.begin() as txn:
raw_data = txn.get(b'00000000')
datum = Datum.FromString(raw_data)
img = np.fromstring(datum.data, dtype=np.uint8)
embed()