On issues encountered in training and evaluating CVTNet on the KITTI dataset

[KuangWenqing]

Hello author, thank you very much for your work. I have encountered two problems in reproducing your work.

Number one: I would like to know whether the CVTNet training on KITTI dataset uses the "train_set.npz" file in the" overlaps "folder under the 03-10 sequence that works the same way as the" OverlapTransformer" and does not sample according to "overlap". All joined in training. If not, how did you do it? When I ran "gen_training_index_kitti.py", I found that there was a problem with repeated sampling, and the final results of "F1" and "AUC" were very different from the ones given in the article.

Second point: evaluation of the "loop closure detection" task on KITTI data 00 sequence. Whether the code and truth values used in the evaluation of "F1" and "AUC" in Tabel 2 are the same as those used in the evaluation of "OverlapTransformer" on KITTI's 00 sequence. The ground truth file called "loop_gt_seq0.00.3overlap_inactiv.npz"?

[allenzg]

我也遇到了同样的问题，我无法复现kitti数据集的相关指标，是否是因为缺少相关数据集的预训练模型，还是因为在具体的gt文件中是用loop_gt_seq0.00.3overlap_inactiv.npz"还是gt_15dis.npy。