Yifan Wang

非常感谢对我们工作的支持和关注！我们的训练代码还在整理中，预计在5月开源。另外，如果想把loftr的backbone换成repvgg后用loftr的代码训练，需要注意backbone和FPN的interpolate的align_corner=True和False要对应不同的supervision计算方式(可以参考[这篇文章](https://ppwwyyxx.com/blog/2021/Where-are-Pixels/))。同时如果直接替换成repvgg会导致较小的精度损失，需要同时使用paper里3.4节的efficient fine feature extraction策略。

非常抱歉训练代码没有准时开源，我们的训练代码现在已经开源，欢迎使用！

Hi Sarlin, Thank you for your interest in our work. We address your questions as follows. 1\) Firstly, we clarify that all statistics in Fig. 1 are from Tab. 1,...

> I think what you're trying to convey with Figure 1 is that you can achieve a AUC@5 of 56 at a speed of 40ms, but that's not actually the...

Thank you very much for the detailed and insightful reply. **1) Whether to Include SP Time:** > Indeed, the speed reported in the LightGlue paper does not include the time...

We agree that “fps cannot be simply obtained by the inverse of latency” and thanks for the comment. Perhaps our Figure 1 could have been more accurately described using "forward...

Thank you for your interest in our work! Our training code is currently being organized and is expected to be open-sourced this month.

We sincerely apologize for the delay in making our training code open-source. It's now available and open for use.

Sorry for the late reply, this is a very good question! For two images w/ intrinsics, you can use the code [here](https://github.com/zju3dv/EfficientLoFTR/blob/68eb9218c1bdd179db9a07ad4771161b198d98b3/src/utils/metrics.py#L182) to use PoseLib to estimate the relative pose....

Hi, thanks for your question. I just tried on a new 3090 server on autodl, re-cloned the latest version of the code, and recreated the environment following the README, but...