Scaling Up Forest Vision with Synthetic Data
- 1Department of Computer Science and Technology, University of Cambridge, 15 JJ Thomson Avenue, Cambridge, CB3 0FD United Kingdom
- 2Clare Hall, University of Cambridge, Herschel Road, Cambridge, CB3 9AL United Kingdom
- 3Department of Plant Sciences, University of Cambridge, Downing Street, Cambridge, CB2 3EA United Kingdom
Abstract
Accurate tree segmentation is a key step in extracting individual tree metrics from forest laser scans, and is essential to understanding ecosystem functions in carbon cycling and beyond. Over the past decade, tree segmentation algorithms have advanced rapidly due to developments in AI. However, existing public 3D forest datasets are not large enough to build robust tree segmentation systems. Motivated by the success of synthetic data in other domains such as self-driving, we investigate whether similar approaches can help with tree segmentation. In place of expensive field data collection and annotation, we use synthetic data during pretraining, and then require only minimal, real forest plot annotation for fine-tuning. We have developed Cambridge Arboreal Modelling Panoptic 3D (CAMP3D), a new synthetic data generation pipeline to do this for forest vision tasks, integrating advances in game engines with physics-based LiDAR simulation. Using CAMP3D, we have produced a comprehensive, diverse, annotated 3D forest dataset on an unprecedented scale. Extensive experiments with a state-of-the-art tree segmentation algorithm and a popular real dataset show that our synthetic data can substantially reduce the need for labelled real data. After fine-tuning on just a single, real, forest plot of less than 0.1 hectare, the pretrained model achieves segmentations that are competitive with a model trained on the full scale real data. We have also identified critical factors for successful use of synthetic data: physics, diversity, and scale, paving the way for more robust 3D forest vision systems in the future. Our CAMP3D pipeline and the resulting dataset are available at https://github.com/yihshe/CAMP3D.git.
Related
- Color naming needs a non-uniform perceptual prior: Individual differences in hue scaling and the vision–language interface
- Vision In SyndrOmic Neurodevelopmental disorders-15q (VISiON-15q) study: eye tracking to supplement clinical examination for detection of cerebral/cortical visual impairment in children with Angelman and Duplication of 15q Syndromes
- Soil–Arthropod Interactions Across Forest Types in a Mediterranean Mountain Landscape
- A Vision-Language Model as a Teacher for Bird Vocalization Detection
- Soil Enzyme Activities and Microbial Nutrient Limitation Between Restored Vegetation and Primary Forest on the Loess Plateau
BibTeX
@article{she2026scaling,
title = {{Scaling Up Forest Vision with Synthetic Data}},
author = {Yihang She and Andrew Blake and David A. Coomes and Srinivasan Keshav},
journal = {International Journal of Computer Vision},
year = {2026},
month = jul,
doi = {10.1007/s11263-026-02923-y},
url = {https://doi.org/10.1007/s11263-026-02923-y},
note = {CC BY},
}
RIS
TY - JOUR
TI - Scaling Up Forest Vision with Synthetic Data
AU - Yihang She
AU - Andrew Blake
AU - David A. Coomes
AU - Srinivasan Keshav
PY - 2026
DA - 2026/07/01
JO - International Journal of Computer Vision
DO - 10.1007/s11263-026-02923-y
UR - https://doi.org/10.1007/s11263-026-02923-y
AB - Accurate tree segmentation is a key step in extracting individual tree metrics from forest laser scans, and is essential to understanding ecosystem functions in carbon cycling and beyond. Over the past decade, tree segmentation algorithms have advanced rapidly due to developments in AI. However, existing public 3D forest datasets are not large enough to build robust tree segmentation systems. Motivated by the success of synthetic data in other domains such as self-driving, we investigate whether similar approaches can help with tree segmentation. In place of expensive field data collection and annotation, we use synthetic data during pretraining, and then require only minimal, real forest plot annotation for fine-tuning. We have developed Cambridge Arboreal Modelling Panoptic 3D (CAMP3D), a new synthetic data generation pipeline to do this for forest vision tasks, integrating advances in game engines with physics-based LiDAR simulation. Using CAMP3D, we have produced a comprehensive, diverse, annotated 3D forest dataset on an unprecedented scale. Extensive experiments with a state-of-the-art tree segmentation algorithm and a popular real dataset show that our synthetic data can substantially reduce the need for labelled real data. After fine-tuning on just a single, real, forest plot of less than 0.1 hectare, the pretrained model achieves segmentations that are competitive with a model trained on the full scale real data. We have also identified critical factors for successful use of synthetic data: physics, diversity, and scale, paving the way for more robust 3D forest vision systems in the future. Our CAMP3D pipeline and the resulting dataset are available at https://github.com/yihshe/CAMP3D.git.
SN - 0920-5691
SN - 1573-1405
ER -