Awesome

CDRL-SA

Update Note

• (24.08.19) The paper has been accepted into Remote Sensing Letters(RSL).

Unsupervised Change Detection Based on Image Reconstruction Loss with Segment Anything

Hyeoncheol Noh, Jingi Ju, Yuhyun Kim, Minwoo Kim, Dong-Geol Choi

Remote Sensing Letters

Getting Started

Dataset download link :

• LEVIR-CD (Our multi-class LEVIR dataset is included in that link.)
• LEVIR-CD_A2B_B2A
• CLCD-CD
• CLCD-CD_A2B_B2A

SAM download link :

• SAM weight

CDRL-SA
    └──datasets
        ├── LEVIR-CD
            ├── val
            ├── test
            └── train
                ├── A
                ├── B
                └── label
        ├── LEVIR-CD_A2B_B2A
            └── train
                ├── A
                └── B
        ├── CLCD-CD
        └── CLCD-CD_A2B_B2A
    └──pretrain_weight
        └── sam_vit_h_4b8939.pth
        

Train

python main.py --root_path ./datasets/ --dataset_name LEVIR-CD --save_name levir

CDRL Difference Map Generate

python test.py --root_path ./datasets/ --dataset_name LEVIR-CD --save_name levir

CDRL-SA Refine Map Generate

python test_sam.py --root_path ./datasets/ --dataset_name LEVIR-CD --save_name levir