[Safe-Net]A Self-Adaptive Feature Extraction Method for Aerial-view Geo-localization

Code for Safe-Net.

Prerequisites

• torch
• torchvision
• numpy
• pyyaml
• tqdm
• scipy
• matplotlib
• pillow

Dataset & Preparation

Download University-1652 upon request and put them under the ./data/ folder. You may use the request template.

Pretrained Vit-S weights

You can download the pretrained Vit-S weights from the following link and put it in the ./models/pretrain_model folder

• Google Driver

Train & Evaluation

Train & Evaluation on University-1652

bash run_train_test_U1652.sh

• You can change the data_dir and test_dir to your own dataset paths in run_train_test_U1652.sh.

TO-DO List

• Support SUES-200 dataset
• Support the evaluation for different levels of distance
• Support ResNet-50 backbone
• Adding the demo of FPM and FAM
• ...

Reference

• University-1652: pdf|code

• LPN: pdf|code

• RK-Net: pdf|code

• FSRA: pdf|code