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

Abstract

We introduce 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, underpinning many downstream tasks such as plant phenotyping, density estimation, and weed control. In foundation-model era, 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 large-scale 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 cross-scene crop segmentation dataset, with 28,406 images across 30+ species and 15 environmental conditions. We build upon a state-of-the-art semantic segmentation architecture ViT-Adapter, enhance it with dynamic upsampling for improved. detail awareness, and train it with a two-stage self-training 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 (99.86% mIoU), and unseen crop varieties (90.09% mIoU), indicating a new state of the art for crop segmentation.