JL1-CD-Trees

Overview

JL1-CD-Trees is a curated subset of the JL1-CD dataset filtered for tree and woodland cover changes. The source dataset covers a diverse range of geographic regions and land cover types across China, including human-induced and natural surface changes. It supports change detection tasks only — no change captions are included.

Dataset Details

• Source: Filtered subset of the JL1-CD dataset (Liu et al., 2025)
• Geographic Coverage: Multiple provinces across China (Shandong, Ningxia, Anhui, Hebei, Hunan and others)
• Temporal Range: Early 2022 to end of 2023

Dataset Splits

• Training: 244
• Validation: 81
• Test: 83

Data Format

Each example contains:

• Image A: Pre-change RGB satellite image
• Image B: Post-change RGB satellite image
• Change Mask: Binary segmentation mask (0=no change, 1=change)

Filtering Criteria

Examples are selected from JL1-CD based on scene content, retaining image pairs containing visible tree or forest cover changes.

Key Characteristics

• Change Coverage:
  • Mean: 5.04% per image
  • Maximum: 48.79%
• Annotation Focus: Binary pixel-level change masks
• Caption Support: None — change detection only
• Object Geometry: Mixed patterns including urban infrastructure, grassland, and tree cover boundaries

Preprocessing

• All images resized to 256×256 pixels for consistency
• Change masks binarized (0=no change, 1=change)
• Bi-temporal image pairs pre-aligned
• Per-channel normalisation using dataset-specific mean and standard deviation statistics

Use Cases

• Remote sensing change detection across diverse geographic regions and land cover types
• Cross-domain transfer learning from forest to mixed land cover scenes
• Benchmarking model generalisation on high-resolution imagery
• Training and evaluating interactive remote sensing agents

Evaluation Metrics

• Per-class IoU: Separate metrics for change and no-change classes
• Mean IoU (mIoU): Average of both class IoUs
• Note: Overall accuracy is not recommended due to class imbalance

Limitations

• No captions: Change detection only — captioning tasks are not supported
• Seasonal and atmospheric variation: Imagery contains variable atmospheric and seasonal conditions which may affect model performance
• Fixed image size: 256×256 pixels
• Resolution: High-resolution imagery (0.5-0.75m/pixel) may not generalise to medium-resolution datasets

Citation

If you use this dataset, please cite:

@article{brock2026forest,
  title={Forest-Chat: Adapting Vision-Language Agents for Interactive Forest Change Analysis},
  author={Brock, James and Zhang, Ce and Anantrasirichai, Nantheera},
  journal={arXiv preprint arXiv:2601.14637},
  year={2026}
}

@article{liu2025jl1,
  title={JL1-CD: A new benchmark for remote sensing change detection and a robust multi-teacher knowledge distillation framework},
  author={Liu, Ziyuan and Zhu, Ruifei and Gao, Long and Zhou, Yuanxiu and Ma, Jingyu and Gu, Yuantao},
  journal={arXiv preprint arXiv:2502.13407},
  year={2025}
}

Paper information available at: https://huggingface.co/papers/2601.14637.

License

MIT License - Academic re-use purpose only

Contact

For questions or issues regarding this dataset, please contact:

• James Brock: james.brock@bristol.ac.uk
• School of Computer Science, University of Bristol