https://github.com/RituYadav92/Large-Scale-Building-Height-Estimation-RSE24-
How High are We? Large-Scale Building Height Estimation Using Sentinel-1 Sar and Sentinel-2 Msi Time Series
https://github.com/RituYadav92/Large-Scale-Building-Height-Estimation-RSE24-
Last synced: 3 months ago
JSON representation
How High are We? Large-Scale Building Height Estimation Using Sentinel-1 Sar and Sentinel-2 Msi Time Series
- Host: GitHub
- URL: https://github.com/RituYadav92/Large-Scale-Building-Height-Estimation-RSE24-
- Owner: RituYadav92
- Created: 2023-11-18T20:08:29.000Z (almost 2 years ago)
- Default Branch: main
- Last Pushed: 2024-04-08T21:57:05.000Z (over 1 year ago)
- Last Synced: 2024-04-08T23:14:56.266Z (over 1 year ago)
- Homepage: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4762421
- Size: 8.79 KB
- Stars: 4
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
## [How High are We? Large-Scale Building Height Estimation Using Sentinel-1 Sar and Sentinel-2 Msi Time Series](https://www.sciencedirect.com/science/article/pii/S0034425724005820)
We propose T-SwinUNet, an advanced DL model for large-scale building height estimation leveraging Sentinel-1 SAR and Sentinel-2 multispectral time series. The model was trained and evaluated on data from the Netherlands, Switzerland, Estonia, and Germany, and its generalizability is evaluated on an out-of-distribution (OOD) test set from ten additional cities from other European countries. T-SwinUNet predicts building height with a Root Mean Square Error (RMSE) of 1.89 m, outperforming state-of-the-art models at 10 m spatial resolution. Its strong generalization to the OOD test set (RMSE of 3.2 m) underscores its potential for low-cost building height estimation across Europe, with future scalability to other regions. Furthermore, the assessment at 100 m resolution reveals that T-SwinUNet (0.29 m RMSE, 0.75 R^2) also outperformed the global building height product GHSL-Built-H R2023A product(0.56 m RMSE and 0.37 R^2).
### 🎉 Manuscript
Remote Sensing of Environment - https://www.sciencedirect.com/science/article/pii/S0034425724005820
Also at 👉 [EGU 2024](https://meetingorganizer.copernicus.org/EGU24/EGU24-4493.html) &
👉 [ESA URBIS 2024](https://urbis24.esa.int/urbis24-agenda/index9f7c.html?page=browseSessions&form_session=71&presentations=hide)
### 🛠️ Setup
create the conda environment via
```bash
conda env create -f environment.yml
```
### 🏋️♂️ Training
Run the python script `train.py` as follows
```bash
python train.py \
--exp_root 'CKPT PATH' \
--config_file './configs/tswin_unet/exp3.yaml' \
--train-df "TRAIN DATA LIST CSV" \
--data_root "TRAIN DATA PATH"
```
### 🚀 Inference
Run the python script `inference.py` as follows
```bash
python predict.py \
--config_file './configs/tswin_unet/exp3.yaml' \
--output_root 'PREDICTION OUTPUT PATH' \
--exp_root 'CKPT PATH' \
--test-df "TEST DATA LIST CSV" \
--data_root "TEST DATA PATH"
```
### 🎉 Dataset
Dataset : [M4Heights ](https://huggingface.co/datasets/Rituxx96x/M4Heights)
Please note that this is not the exact dataset used in the training. However, M4Heights contains the needed Sentinel-1, Sentinel-2 time series and the references for three out of four countries used in this work. We hope the dataset is useful for the task. Please check the dataset instructions before use, and can always contact us for more details.
### 📈 Results
## 🎓 Citation
Please cite our paper:
```bibtex
@article{yadav2025high,
title={How high are we? Large-scale building height estimation at 10 m using Sentinel-1 SAR and Sentinel-2 MSI time series},
author={Yadav, Ritu and Nascetti, Andrea and Ban, Yifang},
journal={Remote Sensing of Environment},
volume={318},
pages={114556},
year={2025},
publisher={Elsevier}
}
```
### 👋 Contact Info.:
Ritu Yadav (email: er.ritu92@gmail.com)