Two doubts on data

[sizhangyu]

After reading your code, I come up with two doubts on the data usage:

1. It seems that the test set was used for tuning the parameters.

2. The original test samples provided by the dataset were not used for evaluation. Instead, the test samples are re-sampled from the filtered test set, where hard samples are not contained.

I am wondering if your results in the paper were generated using these pieces of code.