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

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.

Code and data availability

The paper describes a wheat head segmentation self-training framework using competition data, but contains no public dataset deposit, code release, or model checkpoint with an authors' URL. No qualifying paper-specific public assets are present.

No evidence-backed public reproduction asset is currently recorded.