Fusion Complexity Inversion: Why Simpler Cross View Modules Outperform SSMs and Cross View Attention Transformers for Pasture Biomass Regression

TL;DR

This study reveals that in agricultural biomass estimation with limited data, simpler fusion modules outperform complex attention-based models, emphasizing backbone quality and local modules.

Contribution

It introduces the concept of 'fusion complexity inversion' and systematically evaluates vision models on pasture biomass data, highlighting the importance of backbone pretraining.

Findings

Two-layer gated depthwise convolution outperforms attention transformers and SSMs.

Backbone pretraining scale has a dominant impact on performance.

Metadata training caps maximum achievable R^2 at approximately 0.829.

Abstract

Accurate estimation of pasture biomass from agricultural imagery is critical for sustainable livestock management, yet existing methods are limited by the small, imbalanced, and sparsely annotated datasets typical of real world monitoring. In this study, adaptation of vision foundation models to agricultural regression is systematically evaluated on the CSIRO Pasture Biomass benchmark, a 357 image dual view dataset with laboratory validated, component wise ground truth for five biomass targets, through 17 configurations spanning four backbones (EfficientNet-B3 to DINOv3-ViT-L), five cross view fusion mechanisms, and a 4x2 metadata factorial. A counterintuitive principle, termed "fusion complexity inversion", is uncovered: on scarce agricultural data, a two layer gated depthwise convolution (R^2 = 0.903) outperforms cross view attention transformers (0.833), bidirectional SSMs (0.819), and full Mamba (0.793, below the no fusion baseline). Backbone pretraining scale is found to monotonically dominate all architectural choices, with the DINOv2 -> DINOv3 upgrade alone yielding +5.0 R^2 points. Training only metadata (species, state, and NDVI) is shown to create a universal ceiling at R^2 ~ 0.829, collapsing an 8.4 point fusion spread to 0.1 points. Actionable guidelines for sparse agricultural benchmarks are established: backbone quality should be prioritized over fusion complexity, local modules preferred over global alternatives, and features unavailable at inference excluded.