successhaha / gf-transformer

Cross-Scale Guided Fusion Transformer for Disaster Assessment Using Satellite Imagery

GF-Former

Cross-Scale Guided Fusion Transformer for Disaster Assessment Using Satellite Imagery
https://ieeexplore.ieee.org/document/10197573

Requirements

Install the necessary package with:

-python 3
-pytorch 1.9.0 and torchvision 0.10.0
-apex https://github.com/NVIDA/apex
-opencv-python
-imgaug
-timm

Dataset

Dataset: https://www.xview2.org

Train

Model：we use the segformer as the backbone(mit_b3.pth).
Pretrained model: https://drive.google.com/drive/folders/1GAku0G0iR9DsBxCbfENWMJ27c5lYUeQA

The first stage: CUDA_VISIBLE_DEVICES=0 python train_segformer_loc.py

The second stage: CUDA_VISIBLE_DEVICES=0 python train_segformer_cls.py

Data Processing Techniques

Models trained on different crops sizes from (448, 448) for heavy encoder to (736, 736) for light encoder. Augmentations used for training:

• Flip (often)
• Rotation (often)
• Scale (often)
• Color shifts (rare)
• Clahe / Blur / Noise (rare)
• Saturation / Brightness / Contrast (rare)
• ElasticTransformation (rare)

Inference goes on full image size (1024, 1024) with 4 simple test-time augmentations (original, filp left-right, flip up-down, rotation to 180).

Architecture

Contact the author e-mail：[email protected] or [email protected]