DepthCropSeg++: Scaling a Crop Segmentation Foundation Model With Depth-Labeled Data

TL;DR

DepthCropSeg++ introduces a foundation model for crop segmentation trained on a large cross-scenario dataset using depth-informed pseudo-labels. It uses a ViT-Adapter backbone with FADE dynamic upsampling and a two-stage self-training pipeline to achieve a mean IoU of $93.11\%$ on a diverse test set, outperforming supervised baselines and the segmentation foundation model SAM by large margins. The approach reduces annotation costs by generating high-quality pseudo-labels from monocular depth maps, enabling scalable cross-field deployment. This work demonstrates strong generalization to unseen crops and challenging illumination, canopy density, and field conditions, with practical implications for phenotyping, density estimation, and weed control.

Abstract

DepthCropSeg++: a foundation model for crop segmentation, capable of segmenting different crop species under open in-field environment. Crop segmentation is a fundamental task for modern agriculture, which closely relates to many downstream tasks such as plant phenotyping, density estimation, and weed control. In the era of foundation models, a number of generic large language and vision models have been developed. These models have demonstrated remarkable real world generalization due to significant model capacity and largescale datasets. However, current crop segmentation models mostly learn from limited data due to expensive pixel-level labelling cost, often performing well only under specific crop types or controlled environment. In this work, we follow the vein of our previous work DepthCropSeg, an almost unsupervised approach to crop segmentation, to scale up a cross-species and crossscene crop segmentation dataset, with 28,406 images across 30+ species and 15 environmental conditions. We also build upon a state-of-the-art semantic segmentation architecture ViT-Adapter architecture, enhance it with dynamic upsampling for improved detail awareness, and train the model with a two-stage selftraining pipeline. To systematically validate model performance, we conduct comprehensive experiments to justify the effectiveness and generalization capabilities across multiple crop datasets. Results demonstrate that DepthCropSeg++ achieves 93.11% mIoU on a comprehensive testing set, outperforming both supervised baselines and general-purpose vision foundation models like Segmentation Anything Model (SAM) by significant margins (+0.36% and +48.57% respectively). The model particularly excels in challenging scenarios including night-time environment (86.90% mIoU), high-density canopies (90.09% mIoU), and unseen crop varieties (90.09% mIoU), indicating a new state of the art for crop segmentation.

Figures (14)

• Figure 1: Comparison between specialized and general-purpose crop segmentation models. (a) Specialized models require separate training for each crop variety, while (b) the general-purpose model enables precise segmentation of arbitrary plant species and scenarios using a single strongly generalized model
• Figure 2: Architecture of the baseline model in DepthCropSeg++: ViT-Adapter with FADE upsampling.
• Figure 3: Species distribution and imaging angles in extended datasets
• Figure 4: Examples of full-coverage crop images
• Figure 5: Composition of extended training and test sets