diff --git a/utils/dataloader.py b/utils/dataloader.py
index d6386af..0d66e46 100644
--- a/utils/dataloader.py
+++ b/utils/dataloader.py
@@ -35,8 +35,8 @@ def __init__(self, db_path, ext):
         if self.ext == '.npy':
             self.loader = lambda x: np.load(x)
         else:
-            self.loader = lambda x: np.load(x)['x']
-            # self.loader = lambda x: np.transpose(np.load(x)['x'])
+            # self.loader = lambda x: np.load(x)['x']
+            self.loader = lambda x: np.transpose(np.load(x)['x'])
         if db_path.endswith('.pth'): # Assume a key,value dictionary
             self.db_type = 'pth'
             self.feat_file = torch.load(db_path)
@@ -157,7 +157,7 @@ def __init__(self, opt):
         # self.fc_loader = HybridLoader(self.opt.input_fc_dir, '.npy')
         self.att_loader = HybridLoader(self.opt.image_feat_dir, '.npz')
         self.att2_loader = None
-        if len(self.opt.image_feat_dir2) != 0:
+        if hasattr(self.opt, 'image_feat_dir2') and len(self.opt.image_feat_dir2) != 0:
             self.att2_loader = HybridLoaderv2(self.opt.image_feat_dir, '.npz', self.opt.image_feat_dir2)
         # self.box_loader = HybridLoader(self.opt.input_att_dir, '.npz')['']
 
@@ -265,10 +265,10 @@ def collate_func(self, batch, split):
         max_att_len = max([i[0] for i in num_bbox_batch])
         data['att_feats'] = np.zeros([len(att_batch)*seq_per_img, 100, att_batch[0].shape[1]], dtype = 'float32')
         for i in range(len(att_batch)):
-            data['att_feats'][i*5:(i+1)*5, :att_batch[i].shape[0]] = np.tile(att_batch[i], (5, 1)).reshape(5, att_batch[i].shape[0], att_batch[i].shape[1])
+            data['att_feats'][i*seq_per_img:(i+1)*seq_per_img, :att_batch[i].shape[0]] = np.tile(att_batch[i], (seq_per_img, 1)).reshape(seq_per_img, att_batch[i].shape[0], att_batch[i].shape[1])
         data['att_masks'] = np.zeros(data['att_feats'].shape[:2], dtype='float32')
         for i in range(len(att_batch)):
-            data['att_masks'][i*5:(i+1)*5, :att_batch[i].shape[0]] = np.ones([5, att_batch[i].shape[0]])
+            data['att_masks'][i*seq_per_img:(i+1)*seq_per_img, :att_batch[i].shape[0]] = np.ones([seq_per_img, att_batch[i].shape[0]])
         # set att_masks to None if attention features have same length
         if data['att_masks'].sum() == data['att_masks'].size:
             data['att_masks'] = None