def thread_lmdb_worker(output_lmdb):
lmdb = lmdb_data.lmdb_data(output_lmdb, int(1e11))
lmdb.open_for_write()
while True:
data_fc = queue_fc.get()
is_end = data_fc['is_end']
if is_end == 1:
break
batch_num = data_fc['batch_num']
ids = data_fc['ids']
fc_feats = data_fc['fc_feats']
np_fc_feats = fc_feats.cpu().numpy()
for j in range(batch_num):
np_fc_feat = np_fc_feats[j]
output = io.BytesIO()
# np.savez(output, x=I) # 921788
np.savez_compressed(output, x=np_fc_feat) # 726509
content = output.getvalue()
lmdb.insert(str(ids[j]), content)
lmdb.commit()
lmdb.close()
print('lmdb end')
def __init__(self, opt):
self.opt = opt
self.folder_path = opt.get('folder_path', '')
self.batch_size = opt.get('batch_size', 1)
self.start = opt.get('start', 0)
self.num = opt.get('num', 7500)
self.seq_per_img = 1
self.use_bu_att = opt.get('use_bu_att', False)
self.input_bu = opt.get('input_bu', False)
self.bu_size = opt.get('bu_size', False)
self.bu_feat_size = opt.get('bu_feat_size', False)
if self.use_bu_att:
self.lmdb_bu = lmdb_data.lmdb_data(self.input_bu)
self.lmdb_bu.open_for_read()
self.files = []
self.ids = []
def isImage(f):
supportExt = [
'.jpg', '.JPG', '.JPEG', '.png', '.PNG', '.ppm', '.PPM'
]
for ext in supportExt:
start_idx = f.rfind(ext)
if start_idx >= 0 and start_idx + len(ext) == len(f):
return True
return False
print('start: ', self.start, 'num: ', self.num)
n = 0
for root, dirs, files in os.walk(self.folder_path, topdown=False):
for file in files:
fullpath = os.path.join(self.folder_path, file)
if isImage(fullpath):
if self.num == -1 or (n >= self.start
and n < self.start + self.num):
self.files.append(fullpath)
self.ids.append(file.split('.')[0])
n = n + 1
self.N = len(self.files)
print('DataLoaderRaw found', self.N, ' images')
self.iterator = 0
def __init__(self, opt, is_only_test = False):
self.opt = opt
self.batch_size = opt.batch_size
self.seq_per_img = opt.seq_per_img
self.images_root = opt.images_root
self.image_size = opt.image_size
self.img_padding_max = opt.img_padding_max
self.input_bu = opt.input_bu
self.bu_size = opt.bu_size
self.bu_feat_size = opt.bu_feat_size
self.use_image = opt.use_image
if opt.use_pre_feat:
self.lmdb = lmdb_data.lmdb_data(opt.path_lmdb)
self.lmdb.open_for_read()
print('DataLoader loading json file: ', opt.input_json)
self.info = json.load(open(self.opt.input_json))
self.ix_to_word = self.info['ix_to_word']
self.vocab_size = len(self.ix_to_word)
print('vocab size is', self.vocab_size)
print('DataLoader laoding h5 file: ',opt.input_h5)
self.input_h5 = h5py.File(self.opt.input_h5, 'r')
print('DataLoader laoding bu file: ',opt.input_bu)
self.lmdb_bu = lmdb_data.lmdb_data(opt.input_bu)
self.lmdb_bu.open_for_read()
# if models.is_inception(opt.cnn_model):
# img_mean = [0.5, 0.5, 0.5]
# img_std = [0.5, 0.5, 0.5]
# else:
# img_mean = [0.485, 0.456, 0.406]
# img_std = [0.229, 0.224, 0.225]
#
# vgg_mean = torch.FloatTensor(img_mean).cuda().view(1, 3, 1, 1)
# self.img_mean = vgg_mean.expand(self.batch_size, 3, self.image_size, self.image_size)
#
# vgg_std = torch.FloatTensor(img_std).cuda().view(1, 3, 1, 1)
# self.img_std = vgg_std.expand(self.batch_size, 3, self.image_size, self.image_size)
if models.is_inception(opt.cnn_model):
self.preprocess = trn.Compose([trn.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])])
else:
self.preprocess = trn.Compose([trn.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])
# image_size = self.input_h5['images'].shape
# self.num_images = image_size[0]
# print('image size', image_size)
# print('read %d image'%(self.num_images))
# load in the sequence data
seq_size = self.input_h5['labels'].shape
self.seq_length = seq_size[1]
print('max sequence length in data is', self.seq_length)
self.label_start_ix = self.input_h5['label_start_ix'][:]
self.label_end_ix = self.input_h5['label_end_ix'][:]
# separate out indexes for each of the provided splits
self.split_ix = {'train':[],'val':[],'test':[]}
for ix in range(len(self.info['images'])):
img = self.info['images'][ix]
if img['split'] == 'train':
self.split_ix['train'].append(ix)
elif img['split'] == 'val':
self.split_ix['val'].append(ix)
elif img['split'] == 'test':
self.split_ix['test'].append(ix)
elif opt.train_only == 0:
self.split_ix['train'].append(ix)
random.shuffle(self.split_ix['train'])
random.shuffle(self.split_ix['val'])
random.shuffle(self.split_ix['test'])
print('assigned %d images to split train'%len(self.split_ix['train']))
print('assigned %d images to split val'%len(self.split_ix['val']))
print('assigned %d images to split test'%len(self.split_ix['test']))
self.iterators = {'train':0, 'val':0 ,'test':0}
self.queues = {}
self.flags = {}
self.threads = {}
if not is_only_test:
for split in ['train']:
self.queues[split] = Queue.Queue(maxsize=1)
self.flags[split] = True
thread = threading.Thread(target=get_batch_worker_thread, args=(split, self, opt.batch_size))
thread.setDaemon(True)
thread.start()
self.threads[split] = thread
import io import numpy as np import db.lmdb_data as lmdb_data import cv2 from scipy.misc import imread, imresize import Dic2Obj path_lmdb = '/media/amds/disk/code_cap/bottom-up-attention/output/aic_val_resnet101_fasterRcnn_lmdb' lmdb_bu = lmdb_data.lmdb_data(path_lmdb) lmdb_bu.open_for_read() image_path = '/media/amds/disk/dataset/aic/ai_challenger_caption_validation_20170910/caption_validation_images_20170910/0003a0755539c426ecfc7ed79bc74aeea6be740b.jpg' image_id = '0003a0755539c426ecfc7ed79bc74aeea6be740b.jpg' value = lmdb_bu.get(image_id) data = np.load(io.BytesIO(value)) data_x = data['x'].tolist() boxes = data_x['boxes'] np_boxes = np.array(boxes) img = imread(image_path) # print(img.shape[0]) img_h = float(img.shape[0]) img_w = float(img.shape[1]) print(img_h, img_w) box_size = np.array([img_w, img_h, img_w, img_h]) np_nboxes = np_boxes / box_size