DINOv3 Visual Representations for Blueberry Perception Toward Robotic Harvesting

TL;DR

This paper evaluates DINOv3's effectiveness as a visual perception backbone for blueberry robotic harvesting, highlighting its strengths in segmentation and limitations in detection tasks, guiding future agricultural vision applications.

Contribution

It demonstrates how DINOv3 can serve as a semantic backbone for agricultural tasks, emphasizing the importance of downstream spatial modeling for effective robotic harvesting.

Findings

Segmentation benefits from stable patch-level representations.

Detection performance is limited by target scale variation.

Cluster detection fails due to relational target modeling issues.

Abstract

Vision Foundation Models trained via large-scale self-supervised learning have demonstrated strong generalization in visual perception; however, their practical role and performance limits in agricultural settings remain insufficiently understood. This work evaluates DINOv3 as a frozen backbone for blueberry robotic harvesting-related visual tasks, including fruit and bruise segmentation, as well as fruit and cluster detection. Under a unified protocol with lightweight decoders, segmentation benefits consistently from stable patch-level representations and scales with backbone size. In contrast, detection is constrained by target scale variation, patch discretization, and localization compatibility. The failure of cluster detection highlights limitations in modeling relational targets defined by spatial aggregation. Overall, DINOv3 is best viewed not as an end-to-end task model, but as a semantic backbone whose effectiveness depends on downstream spatial modeling aligned with fruit-scale and aggregation structures, providing guidance for blueberry robotic harvesting. Code and dataset will be available upon acceptance.