huafei-77/TreeCrown-MM
收藏Hugging Face2026-04-07 更新2026-04-12 收录
下载链接：
https://hf-mirror.com/datasets/huafei-77/TreeCrown-MM
下载链接
链接失效反馈官方服务：
资源简介：
---
license: cc-by-nc-4.0
task_categories:
- image-segmentation
- image-to-text
language:
- en
tags:
- remote-sensing
- tree-crown
- segmentation
- caption
- multimodal
- forest
- ecology
- LiDAR
size_categories:
- 10K<n<100K
pretty_name: TreeCrown-MM
---
# TreeCrown-MM
TreeCrown-MM is a multimodal remote sensing dataset for joint tree crown segmentation and caption. It contains 32,787 tiles (1024×1024 px) with ~330K annotated crowns and 285 species, spanning five ecosystem types (Boreal, Tropical, Subtropical, Temperate, Urban-Mixed) across multiple regions worldwide. Each tile provides four aligned modalities: RGB image, height map (LiDAR CHM or pseudo-depth), segmentation mask, and natural-language caption.
The dataset is split into train (26,787), val (1,500), test (1,500), and benchmark (3,000). Split assignments are in `split.json`, and tile-level metadata is in `metadata.csv`.
| File | Size |
|---|---|
| Boreal.tar.gz | 46 GB |
| Tropical.tar.gz | 20 GB |
| Subtropical.tar.gz | 21 GB |
| Temperate.tar.gz | 27 GB |
| Urban-Mixed.tar.gz | 12 GB |
Each archive extracts to four subdirectories: `images/`, `height/`, `masks/`, and `annotations/`.
## Citation
Coming soon.
**License:** CC BY-NC 4.0 · **Contact:** mengjiao@njfu.edu.cn · **GitHub:** https://github.com/huafei18/TreeCrown-MM
提供机构：
huafei-77
