[WACV 2026] AusSmoke meets MultiNatSmoke: a fully-labelled diverse smoke segmentation dataset

---

Overview

The official implement for our WACV 2026 paper AusSmoke meets MultiNatSmoke: a fully-labelled diverse smoke segmentation dataset paper arXiv. It provides:

• A new wildfire smoke segmentation dataset, named MultiNatSmoke, including smoke images around the world.
• Evaluation scripts for state-of-the-art segmentation models on MultiNatSmoke.
• Baseline results and benchmark metrics.

---

Datasets

Our dataset is partially compiled from various existing public datasets, with added segmentation labels. Please ensure you cite the original datasets before downloading or using our dataset. Most datasets are included in this release; however, the Forest Fire dataset requires a separate download. Use the script curate_kaggle_forest_fire.py (included in code/lib folder) for downloading and extracting the images.

Dataset	Link	License
FIgLib	Link	–
Smoke5K	Link	–
SmokeSeg	Link	–
AI-for-Mankind	Link	CC BY-NC-SA 4.0
Firecam	Link	CC BY-NC-SA 4.0
Boreal Forest Fire	Link	CC BY 4.0
D-Fire	Link	CC0 1.0 (Public Domain)
WSDataset	Link	MIT
FireSpot	Link	CC BY 4.0
FESB-MLID	Link	–
Forest Fire	Link	–

Note: Please use the references listed in the Citation section of this repository when citing these datasets.

---

Supported Models

We evaluate the following state-of-the-art segmentation models:

• U-Net (CNN-based)
• DeepLabV3+ (CNN-based)
• SegFormer (transformer-based)
• Mask2Former (transformer-based)
• FoSp (domain-spefic)
• Trans-BVM (domain-spefic)

The FoSp and Trans-BVM implementations follow their respective official repositories

• FoSp: follow instructions and code at LujianYao/FoSp
• Trans-BVM: follow instructions and code at SiyuanYan1/Transmission-BVM

please visit above github repos for evaluating FoSp and Trans-BVM

---

Usage

1. Prepare the MultiNatSmoke dataset

   • Download MultiNatSmoke Dataset

2. Run Forest Fire Dataset Download Script

   python lib/curate_forest_fire.py
   

3. Train a model

   python model_name/train.py
   

   Replace model_name with one of the following:

   • deeplabv3+
   • mask2former
   • oneformer
   • segformer
   • unet

Model performance will be evaluated used IoU, MSE, F1, Precision and Recall.

---

Citation

If you find our work is useful for your research and works, please cite using below BibTeX:

@InProceedings{Li_2026_WACV,
    author    = {Li, Weihao and Zhao, Hongjin and Zhu, Gao and Ji, Ge-Peng and Wilson, Nicholas and Yebra, Marta and Barnes, Nick},
    title     = {AusSmoke meets MultiNatSmoke: a fully-labelled diverse smoke segmentation dataset},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month     = {March},
    year      = {2026},
    pages     = {7996-8006}
}

While using MultiNatSmoke dataset , please also cite below related works, using below BibTeX:

@inproceedings{pornpholkullapat2023firespot,
  title={Firespot: A database for smoke detection in early-stage wildfires},
  author={Pornpholkullapat, Natthaphol and Phankrawee, Warit and Boondet, Peraphat and Thein, Thin Lai Lai and Siharath, Phoummixay and Cruz, Jennifer Dela and Marata, Ken T and Tungpimolrut, Kanokvate and Karnjana, Jessada},
  booktitle={2023 18th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)},
  pages={1--6},
  year={2023},
  organization={IEEE}
}
@article{raita2023combining,
  title={Combining YOLO v5 and transfer learning for smoke-based wildfire detection in boreal forests},
  author={Raita-Hakola, A-M and Rahkonen, S and Suomalainen, J and Markelin, L and Oliveira, R and Hakala, T and Koivum{\"a}ki, N and Honkavaara, E and P{\"o}l{\"o}nen, I},
  journal={The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences},
  volume={48},
  pages={1771--1778},
  year={2023},
  publisher={Copernicus GmbH}
}

@article{de2022automatic,
  title={An automatic fire detection system based on deep convolutional neural networks for low-power, resource-constrained devices},
  author={de Venancio, Pedro Vinicius AB and Lisboa, Adriano C and Barbosa, Adriano V},
  journal={Neural Computing and Applications},
  volume={34},
  number={18},
  pages={15349--15368},
  year={2022},
  publisher={Springer}
}

@article{braovic2017cogent,
  title={Cogent confabulation based expert system for segmentation and classification of natural landscape images},
  author={Braovic, Maja and Stipanicev, Darko and Krstinic, Damir},
  journal={Adv. Electr. Comput. Eng},
  volume={17},
  number={2},
  pages={85--94},
  year={2017}
}

@article{dewangan2022figlib,
  title={FIgLib \& SmokeyNet: Dataset and deep learning model for real-time wildland fire smoke detection},
  author={Dewangan, Anshuman and Pande, Yash and Braun, Hans-Werner and Vernon, Frank and Perez, Ismael and Altintas, Ilkay and Cottrell, Garrison W and Nguyen, Mai H},
  journal={Remote Sensing},
  volume={14},
  number={4},
  pages={1007},
  year={2022},
  publisher={MDPI}
}

@article{govil2020preliminary,
  title={Preliminary results from a wildfire detection system using deep learning on remote camera images},
  author={Govil, Kinshuk and Welch, Morgan L and Ball, J Timothy and Pennypacker, Carlton R},
  journal={Remote Sensing},
  volume={12},
  number={1},
  pages={166},
  year={2020},
  publisher={MDPI}
}

@inproceedings{pornpholkullapat2023firespot,
  title={Firespot: A database for smoke detection in early-stage wildfires},
  author={Pornpholkullapat, Natthaphol and Phankrawee, Warit and Boondet, Peraphat and Thein, Thin Lai Lai and Siharath, Phoummixay and Cruz, Jennifer Dela and Marata, Ken T and Tungpimolrut, Kanokvate and Karnjana, Jessada},
  booktitle={2023 18th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)},
  pages={1--6},
  year={2023},
  organization={IEEE}
}

@article{vu2023ứng,
  title={Ứng dụng trí tuệ nhân tạo để phát hiện bất thường trong giám sát rừng},
  author={Vũ, Quang Vinh and Trần, Công and Anh, Đạt Trần},
  journal={Journal of Science and Technology on Information and Communications},
  volume={1},
  number={4},
  pages={118--124},
  year={2023}
}

@inproceedings{yan2022transmission,
  title={Transmission-guided bayesian generative model for smoke segmentation},
  author={Yan, Siyuan and Zhang, Jing and Barnes, Nick},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={36},
  number={3},
  pages={3009--3017},
  year={2022}
}

@inproceedings{yao2024fosp,
  title={FoSp: Focus and separation network for early smoke segmentation},
  author={Yao, Lujian and Zhao, Haitao and Peng, Jingchao and Wang, Zhongze and Zhao, Kaijie},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={38},
  number={7},
  pages={6621--6629},
  year={2024}
}

@article{zhang2018wildland,
  title={Wildland forest fire smoke detection based on faster R-CNN using synthetic smoke images},
  author={Zhang, Qi-xing and Lin, Gao-hua and Zhang, Yong-ming and Xu, Gao and Wang, Jin-jun},
  journal={Procedia engineering},
  volume={211},
  pages={441--446},
  year={2018},
  publisher={Elsevier}
}

Contact

henryzhao0615@gmail.com