AWMM-100k (All-Weather Multi-Modal dataset)

Existing multimodal image fusion (MMIF) datasets lack comprehensive coverage of adverse weather conditions. To address this limitation, we introduce AWMM-100K, a large-scale benchmark dataset constructed from RoadScene, MSRS, M3FD, and LLVIP, followed by controlled degradation processes to simulate rain, haze, and snow.

In addition, real-world data were captured using a DJI M30T drone equipped with high-resolution visible and thermal cameras. AWMM-100K contains over 187,699 images with weather conditions categorized into light, medium, and heavy intensities. The dataset supports research on multimodal image fusion as well as image restoration tasks such as dehazing, deraining, and desnowing.

Data Overview

Subset	Content	Approx. size
Rain	Rainy scenes (light / medium / heavy intensity)	14.5 GB
Snow	Snowy scenes (light / medium / heavy intensity)	22.5 GB
Haze	Hazy scenes	0.8 GB
Compound Degradation	Haze+Rain, Haze+Snow, Rain+Snow	2.6 GB
All-Weather	Clean all-weather scenes	3.2 GB
Real	Real-world aligned / unaligned pairs	2.9 GB
AWMM-Text	Text captions + text embeddings (npy)	7.8 GB

Total: ~54 GB, ~210,000 files.

Modality / Pairing

Each scene is organized into modality folders. Image fusion relies on paired IR/VI — files share the same numeric index across folders, e.g.:

Rain/Scene_A/Train/gt_ir/1.png     # clean infrared
Rain/Scene_A/Train/gt_vi/1.png     # clean visible
Rain/Scene_A/Train/ir/1.png        # degraded infrared
Rain/Scene_A/Train/vi_xxx/1.png    # degraded visible

gt_* = ground truth (clean); ir / vi_* = degraded inputs; vi_Light/Medium/Heavy denote degradation intensity. Note Scene_D uses flat modality folders (no Train/Test split).

Download (partial download supported)

# 完整下载
hf download Xilai2002/AWMM-100k --repo-type dataset

# 只下某个子集（按需，省流量）
hf download Xilai2002/AWMM-100k --repo-type dataset --include "Rain/*"
hf download Xilai2002/AWMM-100k --repo-type dataset --include "Snow/Scene_A/**"

Python:

from datasets import load_dataset
ds = load_dataset("Xilai2002/AWMM-100k", split="train")

Citation

If you use this dataset, please cite the original AWMM paper:

@article{li2026all,
  title={All-weather multi-modality image fusion: Unified framework and 100k benchmark},
  author={Li, Xilai and Liu, Wuyang and Li, Xiaosong and Zhou, Fuqiang and Li, Huafeng and Nie, Feiping},
  journal={Information Fusion},
  pages={104130},
  year={2026},
  publisher={Elsevier}
}

License

This dataset is released for non-commercial research purposes. Please refer to the original paper (Information Fusion, Elsevier) or contact the authors for the official licensing terms before any commercial use.