Pseudo-Label Refinement for Robust Wheat Head Segmentation via Two-Stage Hybrid Training

TL;DR

This paper presents a self-training framework with a two-stage hybrid strategy and data augmentation for improving wheat head segmentation, achieving competitive results in a global competition.

Contribution

It introduces a novel pseudo-label refinement approach using iterative teacher-student loops combined with hybrid training and extensive data augmentation.

Findings

Achieved competitive performance on development and testing datasets.

Demonstrated effectiveness of iterative pseudo-label refinement.

Validated the robustness of the hybrid training strategy.

Abstract

This extended abstract details our solution for the Global Wheat Full Semantic Segmentation Competition. We developed a systematic self-training framework. This framework combines a two-stage hybrid training strategy with extensive data augmentation. Our core model is SegFormer with a Mix Transformer (MiT-B4) backbone. We employ an iterative teacher-student loop. This loop progressively refines model accuracy. It also maximizes data utilization. Our method achieved competitive performance. This was evident on both the Development and Testing Phase datasets.