search

zhigao2017 / Meta-causal

Code of CVPR 2023 paper Meta-causal Learning for Single Domain Generalization
12 stars 2 forks source link

Is the comparison fair? #1

Open FlyGreyWolf opened 5 months ago

FlyGreyWolf commented 5 months ago

Thanks for your great work! Is it fair to compare when you use 16 data augmentations while other methods don't use as many?

zhigao2017 commented 5 months ago

Thanks for your great work! Is it fair to compare when you use 16 data augmentations while other methods don't use as many?

Thanks for your interest in our method. We follow PDEN, AA, and RA methods to perform data augmentation, and we achieve better results compared with them in our tables. In addition, we argue that data augmentation is one popular scheme for domain generalization, and it is one part of our method (simulate the domain shift). We use the same training data and network architecture with compared methods, thus we think the comparisons are fair.