This repo contains the official implementation of the ACM MM 2024 paper

paper(MM'24)

This code is based on the Sample4Geo framework.

Details of the datasets, training and inference can be found in Sample4Geo.

• 17 Sep 2025 We provide our all model checkpoints on google drive: CVUSA,CVACT,VIGOR (3.06G, including distance_dict, convnext_base, the model with crop, sample, and crop&sample), University-1652 (0.6G, including D2S and S2D).

• 12 Mar 2025 We now support training and testing University-1652 by adding the Class TimmModel_u to mfrgn.py. It is worth noting that the full shared network is used here, which is different from the TimmModel (which only shares the backbone for ground and aerial view). And we provide our pretrained results: checkpoint_u1652 [BaiduYun, Password: 1234].

• 12 Dec 2024 We now provide supplementary results on University-1652.

To accelerate training/test time, you can run data_preparation.py， which implements image transformation (from '.jpg'/'.png' to '.pt') and cropping (similar to SAFA).

When you process images from '.jpg'/'.png' to '.pt', you should set ext='pt' in sample4geo/dataset/*.py

Also, if you are experiencing network errors about the backbone, you may need to download the backbone weights offline and put them into the pretrained folder.

We provide our pretrained results: MFRGN-pretained.zip [BaiduYun, Password: 1234], which contains pretrained weight files or files necessary to train certain network configurations (e.g. distance_dict, convnext backbone weights).