search

amaralibey / Bag-of-Queries

BoQ: A Place is Worth a Bag of learnable Queries (CVPR 2024)
MIT License
97 stars 6 forks source link

About Nordland* and Nordland** results for DINOV2 #11

Closed xjh19971 closed 1 month ago

xjh19971 commented 1 month ago

Hi,

Thank you for the impressive work! I have a question regarding the Nordland* and Nordland* results for DINOV2, as presented in the README file. From the results, it appears that Nordland—which seems to be more challenging—actually achieves better performance than Nordland**, which seems unexpected:

DINOV2 Results:

Nordland*: 81.3 | 92.5 | 94.8 | 96.3 Nordland: 90.6 | 96.0 | 97.5 | 98.5

Could you please confirm whether the Nordland results are correct? Is it possible that the results of the two sets might have been swapped?

Thank you!

xjh19971 commented 1 month ago

Okay, I found that the definition of Nordland* and Nordland** are swapped compared with the paper

amaralibey commented 1 week ago

Okay, I found that the definition of Nordland* and Nordland** are swapped compared with the paper

Thank you @xjh19971, I just updated the README and the ArXiv version.