question

[luckencoder]

The model structure of this paper may be suspected of plagiarizing your feature aggregation method
https://arxiv.org/pdf/2412.00784v1

[amaralibey]

Hello @luckencoder

Thank you for pointing that, the technique has indeed a lot of overlap with BoQ. I will write about this on Twitter to see what the community thinks.

Best,

[Tong-Jin01]

Hello @luckencoder

Thank you for pointing that, the technique has indeed a lot of overlap with BoQ. I will write about this on Twitter to see what the community thinks.

Best,

Hello @amaralibey
BoQ is an outstanding and pioneering work in VPR that has greatly inspired me. In my preprint paper (arXiv:2412.00784v1), I have cited BoQ and highlighted its strengths. In the method section (the last paragraph of Sec III.B), I also emphasize the differences between my approach and BoQ. My structure abandons the encoder and adopts a decoder-only structure for feature aggregation. Furthermore, I use a single set of learnable queries as input, whereas BoQ employs multiple independent queries to process the outputs from each encoder. These differences can help achieve more efficient VPR. Additionally, I am sorry that the claim of being "completely different" is not entirely accurate, and I will make modifications in subsequent versions. About the performance of BoQ in my experiments, I clarified that I re-trained BoQ using 224x224 images (see the caption of Table II) instead of 280x280 images in the original BoQ paper, in order to make a more fair comparison with our method. In summary, the contributions in the paper are my own and are not the same as those in BoQ.