def write_to_lmdb(image_list, db_path, h5_root):
map_size = 150000000000
env = lmdb.Environment(db_path, map_size=map_size)
with env.begin(write=True, buffers=True) as txn:
for idx, image in enumerate(image_list):
# 150 GB
X = get_caffe_data.load_h5_file(os.path.join(h5_root, image[0]))
X = X[np.newaxis, :, :]
y = int(image[1])
datum = array_to_datum(X, y)
str_id = '{:08}'.format(idx)
txn.put(str_id.encode('ascii'), datum.SerializeToString())
env.close()
print " ".join(["Writing to", db_path, "done!"])
# define preprocessing
split = 1
_, test_list = get_caffe_data.get_train_test_lists(label_root, split)
#REMOVE TO WORK:
cur_start = 0
cur_finish = 10
increment = 10
frames = []
net.blobs['data'].reshape(10,1,150,4096)
while cur_finish < len(test_list):
cur_list = test_list[cur_start:cur_finish]
test_data = [get_caffe_data.load_h5_file(os.path.join(h5_root, val[0])) for val in cur_list]
test_data = np.array(test_data, dtype=np.float32)
test_data = np.ascontiguousarray(test_data[:,np.newaxis,:,:])
net.blobs['data'].reshape(10,1,150,4096)
net.blobs['data'].data[:,:,:,:] = test_data
net.forward()
argmax_vals = net.blobs['conv1'].data.argmax(2)
for val in range(10):
cur_counter = Counter(argmax_vals[val,:,0])
best_frame = cur_counter.most_common(1)[0][0]
frames.append(best_frame)
cur_start += increment
cur_finish += increment
final_list = test_list[cur_start:]