Cannot reproduce the reuslts of alphaQE+RRT

[yuyouxixi]

Hello, i tried to reproduce your results shown in the iccv paper, while cannot get the consistent results.

I have checked all the hyperparameters to ensure them to be consistent with those in the paper. Do you have any idea with this?

[fwtan]

Thank you for pointing this out! I confirm that there is an error in Table 3 of our paper: the numbers for QE+RRT were obtained by reranking the top-200, instead of top-100 of the images. These can be verified by changing this line of code. I'm sorry for this mistake. The results do not change the observation made in the paper: combining QE and RRT outperforms QE-only and RRT-only. I will submit a revised version of the paper to arxiv.