serizba / salad

Optimal Transport Aggregation for Visual Place Recognition
GNU General Public License v3.0
140 stars 13 forks source link

Resnet+SALAD #11

Closed YongDeng715 closed 2 weeks ago

YongDeng715 commented 3 months ago

Hello, I would like to reproduce the experiment of Resnet+SALAD in the paper, but cannot find the corresponding setting parameters about training a Resnet+SALAD model. So that I can compare the precision and detecting efficiency of DINOv2 backbone and Resnet backbone. Could you please provide training initialization hyperparameters set of Resnet+SALAD?

serizba commented 3 months ago

Hi @YongDeng715 , when training a ResNet backbone we used the default hyperparameters from MixVPR.