search

zju3dv / GIFT

Code for "GIFT: Learning Transformation-Invariant Dense Visual Descriptors via Group CNNs" NeurIPS 2019
GNU Affero General Public License v3.0
191 stars 17 forks source link

what is "st_peters_12.txt" #16

Open XuYifan98 opened 2 years ago

XuYifan98 commented 2 years ago

Thanks for sharing your code. I try to reproduce your results, I found in the demo there are some .txt files "stpeters.txt". l did not find a way to generate them. I will be appreciated if you inform me how to generate such files for different images.

Thank you in advance.

liuyuan-pal commented 2 years ago

Hi, thanks for your injterest. This file is generated from the YFCC100M dataset, "K" means 3x3 intrinsic matrix and "R" means relative rotation matrix. Evaluation metrics like the AUC used in SuperGlue and LoFTR would be more accurate. https://github.com/zju3dv/LoFTR/blob/94e98b695be18acb43d5d3250f52226a8e36f839/src/utils/metrics.py#L12

szgy66 commented 2 years ago

Hi, thanks for your injterest. This file is generated from the YFCC100M dataset, "K" means 3x3 intrinsic matrix and "R" means relative rotation matrix. Evaluation metrics like the AUC used in SuperGlue and LoFTR would be more accurate. https://github.com/zju3dv/LoFTR/blob/94e98b695be18acb43d5d3250f52226a8e36f839/src/utils/metrics.py#L12 Hello, I also have some confusion about the generation of this .txt file. If I am testing with my own data set, what should my .txt file say, which are camera internal parameters and rotation matrix of two images respectively?