--- license: cc-by-nc-4.0 library_name: pytorch pipeline_tag: image-segmentation datasets: - DailyWildfireCAN/DailyWildfireCAN tags: - wildfire - wildfire-spread-prediction - remote-sensing - spatio-temporal - benchmark - canada arxiv: 2611.14207 --- # ๐Ÿ DailyWildfireCAN Benchmark: Model Checkpoints [๐Ÿ“„ Paper](https://arxiv.org/abs/2611.14207) ยท [๐ŸŒ Project page](https://dailywildfirecan.github.io) ยท [๐Ÿ’ป Code](https://github.com/hagerradi/DailyWildFireCAN) ยท [๐Ÿ—‚๏ธ Dataset](https://huggingface.co/datasets/DailyWildfireCAN/DailyWildfireCAN) This repository contains the trained checkpoints for the models in the **DailyWildfireCAN** benchmark for next-day wildfire spread prediction across Canada. Each model was trained three times with different random seeds, and all three runs are released. ![DailyWildfireCAN benchmark](https://cdn-uploads.huggingface.co/production/uploads/69f8c2efff7c500b6101eddd/kZlcdvFOCwqimxEmgMZFW.png) ## Task Given the inputs available up to day *T*, predict which pixels will newly burn on day *T+1*. - **Input:** a 256 ร— 256 tile at 90 m resolution with 50 channels (see the [dataset card](https://huggingface.co/datasets/DailyWildfireCAN/DailyWildfireCAN) for the channel order). Multi-temporal models take 3 consecutive days: (3, 50, 256, 256). - **Output:** one logit per pixel, (1, 256, 256). Apply a sigmoid to get the probability of new burn. - **Metrics:** AUC-PR, precision and recall, computed on the Test-Easy (unseen year), Test-Medium (unseen regions) and Test-Hard (unseen regions and year) splits. ## Models and results AUC-PR is the mean over three seeds. Full results with precision, recall and standard deviations are in Tables 5 and 6 of the [paper](https://arxiv.org/abs/2611.14207). ### Mono-temporal models (single day of input) | Model | Folder | `architecture` in code | Size / run | Test-Easy | Test-Medium | Test-Hard | |---|---|---|---|---|---|---| | U-Net | `unet` | `unet` | 373 MB | 0.285 | 0.218 | 0.175 | | U-Net + Time-Gap Encoding | `unet_with_time_gap` | `unet_time_gap` | 374 MB | **0.287** | **0.233** | 0.192 | | Attention U-Net | `attention_unet` | `unet_attention` | 377 MB | 0.280 | 0.199 | 0.155 | | SegFormer U-Net | `segformer_unet` | `unet_segformer` | 571 MB | 0.205 | 0.181 | 0.162 | | Swin-U-Net | `swin_unet` | `asufm` | 413 MB | 0.185 | 0.148 | 0.102 | | OlmoEarth | `olmoearth` | `unet_olmo_offline` | 458 MB | 0.277 | 0.221 | 0.183 | | DINOv3 | `dinov3` | `dual_dinov3` | 1.7 GB | 0.238 | 0.194 | **0.189** | ### Multi-temporal models (3 consecutive days of input) | Model | Folder | `architecture` in code | Size / run | Test-Easy | Test-Medium | Test-Hard | |---|---|---|---|---|---|---| | U-TAE | `utae` | `utae` | 139 MB | **0.352** | **0.306** | **0.259** | | U-Net ConvLSTM | `unet_with_convlstm` | `unet_convlstm` | 348 MB | 0.340 | 0.278 | 0.222 | | SimVPv2 | `simvpv2` | `simvpv2_spatiotemporal` | 66 MB | 0.284 | 0.250 | 0.203 | | Video Swin-U-Net | `video_swin_unet` | `video_swin_unet` | 539 MB | 0.272 | 0.235 | 0.178 | | Multi-Modal | `multi_mode_model` | `frp` | 4 MB | 0.182 | 0.167 | 0.110 | Model descriptions and references are in the [code repository](https://github.com/hagerradi/DailyWildFireCAN#-part-4-modeling). ## Repository structure ``` benchmark_models_checkpoints/ โ”œโ”€โ”€ unet/ โ”‚ โ”œโ”€โ”€ checkpoints_unet_run_1/best_checkpoint.pt โ”‚ โ”œโ”€โ”€ checkpoints_unet_run_2/best_checkpoint.pt โ”‚ โ””โ”€โ”€ checkpoints_unet_run_3/best_checkpoint.pt โ”œโ”€โ”€ unet_with_time_gap/ โ”‚ โ””โ”€โ”€ ... โ””โ”€โ”€ /checkpoints__run_/best_checkpoint.pt ``` The full repository is about 16 GB. `best_checkpoint.pt` is the checkpoint with the best validation score. It is a dictionary with the keys `model_state_dict`, `optimizer_state_dict`, `scheduler_state_dict`, `epoch` and `best_val_loss`. ## Quick start Clone the code repository, which contains the model definitions: ```bash git clone https://github.com/hagerradi/DailyWildFireCAN cd DailyWildFireCAN pip install -r requirements.txt ``` Download a checkpoint and run inference: ```python import torch from huggingface_hub import hf_hub_download from src.models.unet import UNet path = hf_hub_download( repo_id="DailyWildfireCAN/DailyWildfireCAN-Benchmark", filename="benchmark_models_checkpoints/unet/checkpoints_unet_run_1/best_checkpoint.pt", ) model = UNet(input_channels=50, num_classes=1, hidden_features=[64, 128, 256, 512]) ckpt = torch.load(path, map_location="cpu") model.load_state_dict(ckpt["model_state_dict"]) model.eval() x = torch.randn(1, 50, 256, 256) # replace with a normalized sample's "x" with torch.no_grad(): prob = torch.sigmoid(model(x)) # (1, 1, 256, 256) probability of new burn ``` For the other models, build the model the same way as in `src/evaluate.py` with the settings in `configs/default.yaml`, then load `model_state_dict`. Download all runs of one model: ```python from huggingface_hub import snapshot_download snapshot_download( repo_id="DailyWildfireCAN/DailyWildfireCAN-Benchmark", allow_patterns=["benchmark_models_checkpoints/utae/*"], local_dir="checkpoints", ) ``` ## Training details All models were trained with the same setup, using the full set of input features and the training-set normalization statistics released with the dataset. See Appendix B of the [paper](https://arxiv.org/abs/2611.14207) for hyperparameters and hardware. ## License The checkpoints are released under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license. The models were trained on [DailyWildfireCAN](https://huggingface.co/datasets/DailyWildfireCAN/DailyWildfireCAN), which is released under CC BY-NC-SA 4.0 and includes third-party data subject to their own licenses. ## Citation ```bibtex @article{dailywildfirecan, title={DailyWildfireCAN: A Dataset and Benchmark for Next-Day Wildfire Spread Prediction in Canada}, author={Ettahri, Mahacine and Abdelwahed, Hager Radi and Olutayo, Toluwaleke and Aloise, Daniel and Caron, Ga{\'e}tan Marceau}, journal={arXiv preprint arXiv:2611.14207}, year={2026} } ``` ## Contact For inquiries, contact Mahacine Ettahri (mahacine.ettahri@etud.polymtl.ca) or Hager Radi (hager.radi@mila.quebec).