This repository contains the fine-tuned weights for a UNet with resnext50_32x4d backbone (default architecture of Microsoft's damage-building-assessment repository), to perform building damage assessment. The model takes an RGB image as input and outputs a segmented image where each pixel represents one of three classes: background, healthy building, or damaged building (example below).

Follow the steps below to use the fine-tuned weights and perform building damage assessment on your custom images:

Download the fine-tuned weights from this repo

Clone the damage-building-assessment repository.

Create a new config file for your experiment in the config folder and run the script project_setup.py

Add the fine-tuned weights in the experiments/your-experiment/checkpoints folder

run the script inference.py on your custom images.

More instructuons are provided here