train.py RuntimeError

when I train on my own data,I meet an error when run the faster_rcnn.py,line 100,'rpn_loss_box = F.smooth_l1_loss(rpn_bbox_pred, rpn_bbox_targets, size_average=False) / (fg_cnt + 1e-4)'. It shows that "RuntimeError: Expected object of type torch.cuda.FloatTensor but found type torch.cuda.LongTensor for argument #2 'other'“
what's mabey the reason??