Eval-MegaDepth-1500

[zw-92]

Hello, I evaluated the method on MegaDepth based on the evaluation of LOFTR, but the effect is not good, I can't figure out the problem for the time being. Looking forward to your assessment.

[guipotje]

Hello @zw-92, thank you for the interest in our work!

Please notice that we use poselib in our paper's evaluation, which provides much better and faster pose estimation than the opencv default implementation used in LoFTR

[Wang86760]

@zw-92 你好，可以分享一下您用LOFTR评估方法评估Xfeat的代码吗？谢谢

[cw314]

@zw-92 Hello, can you provide the code for evaluating the Xfeat model?

[zw-92]

不好意思，之前有事耽误了。之前因为评估指标非常低，所以暂时将该评估工作放下了。最近因为算力原因，又开始重新研究，等再次核验代码后，会提供评估代码。

[zw-92]

Hello @zw-92, thank you for the interest in our work!

Please notice that we use poselib in our paper's evaluation, which provides much better and faster pose estimation than the opencv default implementation used in LoFTR

[zw-92]

Hello @zw-92, thank you for the interest in our work!

Please notice that we use poselib in our paper's evaluation, which provides much better and faster pose estimation than the opencv default implementation used in LoFTR

Hi, your work is very good, I have reproduced the accuracy of the method in the MegaDepth 1500, and even improved by 1-2 points. I was wondering if I could replace ALike with ROMA if I wanted to optimize XFeat+LightGule.

[Wang86760]

@zw-92 你好，可以分享一下怎么使用的poselib吗，我尝试了很多方法都没能实现评估

[xushangnjlh]

anyone reproduce the AUC on ScanNet? I have reproduced AUC on MegaDepth1500, but ScanNet AUC is not that high...

[Odin-byte]

@xushangnjlh Hey, I am also observing this behavior. Where you able to spot an error in you approach? My scannet evaluation also does not reach the exspected values.