ConvNeXt or ConvNeXt-V2 for drone target view geo-localization problem ?

[blackpearl1022]

Thanks for your amazing work !
I have been working on drone target view geo-localization using Sample4Geo and noticed they are using ConvNeXt backbone for their approach.
I used the ConvNeXt-V2 for same approach using the rest of the training codebase without any modification and trained with my own dataset which is similar with University-1652 dataset, but after training I realized that the result with ConvNeXt-V2 is much mess than ConvNeXt even though the pre-trained model of ConvNeXt-V2 shows the better performance. And also in speed side, ConvNeXt much better than ConvNeXt-V2.
What's wrong with this ?