University of Angers: Agricultural AI Data Annotation for Apple Tree Phenology
University of Angers
Industry
Agricultural
Type of service
Bounding box annotation
Platform used
Results
Impact
The client
The University of Angers, through its LARIS laboratory and IRHS research institute, conducts research in artificial intelligence, computer vision, and precision agriculture. Their work focuses on developing AI tools for phenology analysis in agricultural environments, in collaboration with agronomists as part of the PHENET European project.
The challenge
The team’s research required annotating large volumes of images containing apples, fruitlets, and flowering stages, collected across multiple orchards and countries. These annotations needed to be highly consistent, and the University of Angers team needed a reliable external partner who could deliver high-quality work within research timelines.
Plant and fruit focused annotation is inherently complex, objects vary across different stages of growth, and samples can look markedly different from one orchard to the next.
The work
The project involved annotating 225,110 objects representing different phenological stages of apple trees under varying environmental conditions.
Humans in the Loop delivered bounding box annotations according to the research team’s guidelines. Throughout the project, the two teams worked closely to clarify requirements, address feedback, and refine the annotation workflow as needed, helping maintain consistency across the dataset.
The result
The annotated dataset contributed directly to DeepPhenoTree, a multi-site, RGB-annotated plant phenology image dataset built for deep learning applications in horticulture. The associated data paper provides further details on the dataset and research methodology.
The dataset follows an entire growing season across a pan-European experimental network of apple orchards. It includes a large set of shared genotypes grown in different environments, giving the resulting AI models a wide range of image contrasts to learn from and contributing to their robustness.
Apple Flowers with Bounding Box Annotations
Mature Apples with Bounding Box Annotations
The dataset has also directly enabled:
- A public data challenge, hosted on Codabench and promoted through the CVPPA 2026 workshop at the European Computer Vision Conference.
- A standalone, open-source application built around an AI model trained on the annotated dataset, released on the Napari platform.
With this tool, breeders, farmers, technical institutes, and agtech providers can track key phenological stages, detect flowering peaks, estimate harvest timing, and guide robots for pruning.
“The whole process was very smooth and reactive. Annotation quality is high and reliable. We would collaborate again with HITL for another round of our data challenge around the DeepPhenoTree dataset.”
David Rousseau, Distinguished professor, Université d’Angers
Social Impact
A total of 10 conflict-affected annotators contributed to this project. Eight annotators were based in Lebanon, including four men and four women, while two annotators, both women, currently residing in Bulgaria, also took part.
At Humans in the Loop, providing dignified opportunities for people affected by conflict isn’t just part of our mission – it’s central to how we produce high‑quality, inclusive, and ethical data for AI.
Interested in implementing a human-in-the-loop in your agricultural AI research? Get in touch with us and we’d be happy to have a call.
