DAHiTra: UNET Architecture with Hierarchical Transformers for Automated Building Damage Assessment Using Satellite Imagery

A novel transformer-based network model is presented for building damage assessment which leverages hierarchical spatial features of multiple resolutions and captures temporal difference in feature domain after applying transformer encoder on the spatial features. The proposed network achieves state of the art performance while tested on a large-scale disaster damage dataset (xBD) for building localization and damage classification, as well as on LEVIR-CD dataset for change detection tasks.

Publication accepted in CACAIE (peer-reviewed journal): https://onlinelibrary.wiley.com/doi/10.1111/mice.12981

Arxiv link: https://arxiv.org/abs/2208.02205

Results

Average Quantitative Results for Damage Classification: xBD dataset

Average Quantitative Results for Change Detection: LEVIR-CD dataset

Weights

Model trained on xBD dataset for 120 epochs. Weights
See logs here Log file

Training the model

Requirements

Python 3.6
pytorch 1.6.0
torchvision 0.7.0
einops  0.3.0

Installation

Clone this repo:

git clone https://github.com/nka77/DamageAssessment.git
cd DamageAssessment

Training the model

Please refer the training script run_cd.sh and the evaluation script eval.sh in the folder scripts.

Training goal specific files:

1. Classification: xBD_code/train.py
2. Localization: xBD_code/train_loc.py
3. Domain adaptation: xBD_code/train_adapt.py

Dataset Preparation

Data Download

1. xBD: https://xview2.org/dataset
2. LEVIR-CD: https://justchenhao.github.io/LEVIR/

Required data format

"""
xBD damage classification data set with pixel-level binary labels；
├─train
    |-images
    ├─masks
├─tier3
    |-images
    ├─masks
├─test
    |-images
"""

train and tier3 : pre-disaster and post-disaster images; masks: 5 class label maps;

"""
LEVIR Change detection data set with pixel-level binary labels；
├─A
├─B
├─label
└─list
"""

A: images of t1 phase; B:images of t2 phase; label: label maps; list: contains train.txt, val.txt and test.txt, each file records the image names (XXX.png) in the change detection dataset.

Acknowledgements

We thank https://github.com/justchenhao/BIT_CD.git for providing the code base publicly. We develop our code on the top of this repository.