AgriField-40K: Adapting Vision Models to Agriculture With Efficient Continual Pretraining

2026-08-08

AI Summary

This work addresses the challenges of high annotation costs and substantial adaptation overhead in agricultural vision tasks, which stem from the lack of efficient pretraining strategies. To overcome these limitations, the authors propose AgriMAE, a method that builds upon a masked autoencoder pretrained on natural images and introduces lightweight adapters for domain-specific continual pretraining on agricultural data. Notably, AgriMAE is the first to incorporate semantic feature reconstruction as a pretraining objective. Evaluated on AgriField-40K—a newly curated dataset of 40,000 in-field agricultural images—AgriMAE demonstrates significant performance gains across multiple downstream tasks. It achieves comparable or superior results to full fine-tuning while requiring up to nine times fewer trainable parameters, thereby validating its efficiency and practical utility in real-world agricultural applications.

Abstract

Field-based agricultural computer vision is important for precision agriculture, yet it largely depends on expensive annotations and costly adaptation of large pretrained models. We introduce AgriField-40K, a field-centric dataset curated from 17 public resources and covering diverse crops, weeds, and field conditions. Building on this, we present AgriMAE, a parameter-efficient continual pretraining baseline that adapts a masked autoencoder pretrained on natural images by training only lightweight adapters. We further explore semantic feature reconstruction as an alternative pretraining objective and evaluate transfer across multiple tasks. AgriMAE consistently improves downstream performance and can match or even outperform full fine-tuning while using up to $9\times$ fewer trainable parameters, showing that AgriField-40K is a practical resource for continual pretraining in agricultural vision. Project page: https://dtu-pas.github.io/agrifield40k/

Problem

• agricultural computer vision
• precision agriculture
• expensive annotations
• model adaptation
• field-based vision

Innovation

• continual pretraining
• parameter-efficient adaptation
• masked autoencoder
• agricultural computer vision
• adapter-based fine-tuning