A Root Foundation Model for Zero-Shot Segmentation

Foundation models pre-trained on massive datasets have demonstrated impressive performance, but in some specialised domains have been found to have lower accuracy. Domain-specific foundation models target a particular domain such as retinal or plant images. These domain-specific models have shown inconsistent results and the benefit to root segmentation is unknown. We train and evaluate the first domain-specific foundation model for root segmentation. Evaluation uses a leave-one-dataset-out design across nine diverse root datasets with two architectures. The domain-specific model segments unseen root datasets zero-shot (without any fine-tuning on the unseen dataset), achieving a mean Dice of 0.636 versus 0.698 for individually fine-tuned models, that is, 92% of fine-tuned Dice on average and above 90% for 5 of 9 datasets. We also test few-shot transfer learning. Fine-tuning on only 10 patches, the domain-specific model recovers 95% of its full-data Dice on average, versus 69% for a general pre-trained model. With full target-data fine-tuning, the two perform comparably, with mean improvements of +0.011 Dice for MobileSAM and +0.022 for M2F Swin-S, neither significant (Wilcoxon p = 0.150 and 0.064). We release our pre-trained MobileSAM root foundation model for use with RootPainter, enabling fully automatic root segmentation on new datasets with an ordinary laptop or desktop computer, with no need for annotation or training.