training details and trained model

[a-flying-crow]

Hi, i wonder whether the parameters in the code are the same as those used in the result of your paper? If not, could you give me a copy of the experimental parameters? I used the current parameters and tried to adjust the parameters of the third step, and found that the experimental results were a big gap with the paper.

In addition, for finetuning the parameters, does the parameter adjustment process in the first part have much influence on the subsequent results? If so, could you tell me the approximate validation score in the first step?

Lastly, could you please give me the model parameter file corresponding to the experimental results in your paper?
Thanks for your reading.

[Yu-Wu]

@a-flying-crow Thanks for the interest on our paper.

• Are the parameters in the code are the same as those used in the result of your paper?
  I did not change the hyper-parameters. This should be the same to the paper version.

• Does the parameter adjustment process in the first part have much influence on the subsequent results?
  I don't train different parameters in the first part. But I think it would not be too much difference for the final performance.

• Could you please give me the model parameter file corresponding to the experimental results in your paper?
  Sorry that I don't have the copy of the previous model weights.

Could you download the code again and run it following the instructions? I think it reproduce the original results in the paper.