Evaluate the code

[Wangsw02]

Hello, thanks for your work, could you please provide the evaluation code for hpatches and Aachen day-night datasets?

[xtcpete]

Hi, thanks for your interest. I’m not planning to include them in this repo since I don't have much time to do it. For evaluation on HPatches, you could refer to ALIKE. For the Aachen Day-Night dataset, you could refer to this example from Hloc.

[Wangsw02]

Hello, thank you for your answer. I would like to ask: in the HPatches evaluation, does your matching method use the MNN matcher from ALike or the match/match_dense from rdd_helper? The MHA results I obtained from reproducing V1 are 45.93/57.22/63.52, which differ significantly from the results in the paper.

[xtcpete]

Hi, we used the dual softmax matcher in our repo and we followed the protocol described in xfeat which might not be exactly the same as it in ALIKE. Make sure you handle the image resizing correctly and use correct thresholds. I also felt that this evolution on HPatches is saturated. If I had more time before submission I would change it to AUC as described in many other papers like LoFTR. Here is a great tool you can use for running both HPatches and Aachen Day-Night. I hope this can help.