Absolute position in the local branch

[alina2204]

Hello,

Congrats for your work and thanks for sharing the code! I have two questions:

• For computing the absolute position of each feature vector in the local branch, it seems to me that there are some inconsistencies between equations (6) and (7) from the paper and what is implemented in your code. In your code (function _location_to_NxN_grid in Dataset/transforms.py) I think that you implemented l_1 = i_2 + w_2 / (2W) + w_2/W (u-1) instead of Eq (6), which indeed seems to be the correct operation to estimate the feature coordinates in the original image. Can you please check and let me know if I understood correctly ?
• Also, in Dataset/transforms.py line 366, shouldn't it be view_mim_img=view_mim_img instead of view_mim_img=view ?

Thank you in advance,

Alina

[pUmpKin-Co]

Hi~Glad to see that our work will help you!

• So much thanks for you. You're absolutely right about that. When I was writing, I incorrectly assumed that the distance between grid cells wouldn't change after resize, but I just did the calculation again and found that it actually does! Please refer to the code implementation. Thanks.
• Thanks for yours corrections. I have changed the corresponding code. Thanks again for your carefulness in pointing out these errors!