[Paper Review] FOR-instance: a UAV laser scanning benchmark dataset for semantic and instance segmentation of individual trees

Stefano Puliti, Grant D. Pearse+5 more|arXiv (Cornell University)|Sep 3, 2023

Remote Sensing and LiDAR ApplicationsEnvironmental Science26 citations

TL;DR

The FOR-instance dataset provides five curated UAV LiDAR collections with manual annotations for semantic and instance segmentation of individual trees, enabling train/test benchmarking and including DBH data for 3D forest analysis.

ABSTRACT

The FOR-instance dataset (available at https://doi.org/10.5281/zenodo.8287792) addresses the challenge of accurate individual tree segmentation from laser scanning data, crucial for understanding forest ecosystems and sustainable management. Despite the growing need for detailed tree data, automating segmentation and tracking scientific progress remains difficult. Existing methodologies often overfit small datasets and lack comparability, limiting their applicability. Amid the progress triggered by the emergence of deep learning methodologies, standardized benchmarking assumes paramount importance in these research domains. This data paper introduces a benchmarking dataset for dense airborne laser scanning data, aimed at advancing instance and semantic segmentation techniques and promoting progress in 3D forest scene segmentation. The FOR-instance dataset comprises five curated and ML-ready UAV-based laser scanning data collections from diverse global locations, representing various forest types. The laser scanning data were manually annotated into individual trees (instances) and different semantic classes (e.g. stem, woody branches, live branches, terrain, low vegetation). The dataset is divided into development and test subsets, enabling method advancement and evaluation, with specific guidelines for utilization. It supports instance and semantic segmentation, offering adaptability to deep learning frameworks and diverse segmentation strategies, while the inclusion of diameter at breast height data expands its utility to the measurement of a classic tree variable. In conclusion, the FOR-instance dataset contributes to filling a gap in the 3D forest research, enhancing the development and benchmarking of segmentation algorithms for dense airborne laser scanning data.

Motivation & Objective

• Address the need for accurate individual tree segmentation from laser scanning data.
• Provide a standardized, ML-ready benchmark to advance instance and semantic segmentation in dense 3D forest scenes.
• Offer diverse UAV-based LiDAR collections from global locations to support generalizable benchmarking.
• Enable evaluation protocols with development and test splits and practical guidelines for utilization.
• Include diameter at breast height (DBH) data to extend utility to classic tree measurements.

Proposed method

• Curate five UAV-based laser scanning datasets from diverse forest types.
• Manually annotate data into individual tree instances and semantic classes (e.g., stem, woody branches, live branches, terrain, low vegetation).
• Split datasets into development and test subsets with usage guidelines.
• Ensure dataset is adaptable to deep learning frameworks and various segmentation strategies (semantic and instance).
• Provide ML-ready data to promote benchmarking and progress in 3D forest scene segmentation.

Experimental results

Research questions

• RQ1 RQ1 Can UAV-based LiDAR data be effectively annotated for reliable instance segmentation of individual trees?
• RQ2 RQ2 How does the FOR-instance dataset support evaluation of semantic and instance segmentation across diverse forest types?
• RQ3 RQ3 Can deep learning frameworks be effectively applied to this dataset for dense 3D forest scene segmentation?
• RQ4 RQ4 How does annotating DBH data enhance tree-level analysis and segmentation utility?

Key findings

• The FOR-instance dataset comprises five curated UAV-based laser scanning collections suitable for ML-driven segmentation.
• Data are manually annotated into per-tree instances and semantic classes such as stem, woody branches, live branches, terrain, and low vegetation.
• The dataset is divided into development and test subsets to enable method advancement and objective evaluation.
• It is designed to be adaptable to various deep learning segmentation strategies and frameworks.
• Inclusion of DBH data broadens the dataset’s utility for traditional tree measurements alongside segmentation tasks.