What does this research mean for the field?

An adapted SAM3 model combined with 3D Gaussian scene representation, cross-view contrastive learning, and Hough voting significantly improves 3D fruit detection and instance segmentation in densely occluded agricultural environments compared to existing methods like FruitNeRF. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To improve 3D fruit detection, localization, and counting under dense occlusions and variable lighting by adapting a vision foundation model and 3D Gaussian scene representations.

September 2, 2026Computers and Electronics in AgricultureOpen Access

Adapting SAM3 for 3D fruit counting with cross-view contrastive learning and Hough voting

View Full Paper

Ask AI

Bookmark

Authors

KZKai Zhao CKChenchen Kang SRSuzy Rogiers

Discussion

Member takes

Overview

Validation study demonstrates accurate 3D fruit localization across dense orchards, indicating improved capabilities for automated harvesting robots.

Key Points

To improve 3D fruit detection, localization, and counting under dense occlusions and variable lighting by adapting a vision foundation model and 3D Gaussian scene representations.
Adapted SAM3 using low-rank adaptation to produce high-quality 2D instance masks across varying crops and illumination conditions.
Integrated cross-view contrastive learning with uncertainty weighting to mitigate noisy pseudo-labels within 3D Gaussian representations.
Applied 3D Hough voting and clustering to fuse multi-view evidence for counting and localizing fruits across synthetic orchards, real apple orchards, and greenhouse sweet pepper scenes.
Achieved F1-scores of 0.971 for plum and 0.977 for mango in dense, highly occluded orchard environments.
Delivered more than a twofold improvement in 3D instance segmentation performance compared to FruitNeRF.
Demonstrated direct cross-domain transferability from open-field apple orchards to greenhouse sweet peppers by updating only the text prompt.

Cite This Study

Zhao et al. (2026) studied this question.

synapsesocial.com/papers/6a97e249c562ede874ec65bd https://doi.org/10.1016/j.compag.2026.112325

View Full Paper

Ask AI

Bookmark

September 2, 2026Computers and Electronics in AgricultureOpen Access

Adapting SAM3 for 3D fruit counting with cross-view contrastive learning and Hough voting

View Full Paper

Ask AI

Bookmark

Authors

KZKai Zhao CKChenchen Kang SRSuzy Rogiers

Discussion

Member takes

Overview

Validation study demonstrates accurate 3D fruit localization across dense orchards, indicating improved capabilities for automated harvesting robots.

Key Points

To improve 3D fruit detection, localization, and counting under dense occlusions and variable lighting by adapting a vision foundation model and 3D Gaussian scene representations.
Adapted SAM3 using low-rank adaptation to produce high-quality 2D instance masks across varying crops and illumination conditions.
Integrated cross-view contrastive learning with uncertainty weighting to mitigate noisy pseudo-labels within 3D Gaussian representations.
Applied 3D Hough voting and clustering to fuse multi-view evidence for counting and localizing fruits across synthetic orchards, real apple orchards, and greenhouse sweet pepper scenes.
Achieved F1-scores of 0.971 for plum and 0.977 for mango in dense, highly occluded orchard environments.
Delivered more than a twofold improvement in 3D instance segmentation performance compared to FruitNeRF.
Demonstrated direct cross-domain transferability from open-field apple orchards to greenhouse sweet peppers by updating only the text prompt.

Cite This Study

Zhao et al. (2026) studied this question.

synapsesocial.com/papers/6a97e249c562ede874ec65bd https://doi.org/10.1016/j.compag.2026.112325

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

View Full Paper

Ask AI

Bookmark