We show the preprocessing, training and testing time of the models in Table 1. BinPro and B2SFinder also need to use traditional matching algorithms to compute the similarity scores. In comparison with the pre-training models, the end-to-end models get rid of the pre-training time. Compared with random sampling, our norm weighted sampling method requires more time. However, the additional time consumption is worth, because the effect has improved a lot.

We show the preprocessing, training and testing time of the models in Table 1. BinPro and B2SFinder also need to use traditional matching algorithms to compute the similarity scores. In comparison with the pre-training models, the end-to-end models get rid of the pre-training time. Compared with random sampling, our norm weighted sampling method requires more time. However, the additional time consumption is worth, because the effect has improved a lot.