Egomaze dataset aids first-person maize plant segmentation under heavy occlusion
EgoMaize: A First-Person Maize Instance Segmentation Benchmark under Severe Field Occlusion
Computer Vision and Pattern Recognition
Summary
It’s hard to identify and separate individual maize plants in close-up images taken inside dense fields because the plants overlap a lot and look similar. The authors created EgoMaize, a new collection of annotated images that help train and test computer vision models to recognize and separate these plants and their parts, even when they’re heavily hidden behind leaves or other plants. Their study shows that while some techniques help with certain challenges, no current method perfectly solves all aspects of separating overlapping plant parts in these conditions. The dataset and code are available online for others to use and improve upon.
What this means in practice
- •For agriculture technology developers: Improve mobile tools that identify and analyze individual maize plants from close-up field images despite heavy overlap and occlusion.
- •For automated crop monitoring teams: Enhance algorithms that track plant growth by accurately segmenting maize stems and tassels in dense canopy images taken at ground level.
Authors
Jiayi Li, Zihan Zhang, Erhankang Yan, Yitian Chen, Yuze Li, Chengzhang Ding, Jianxin Cao
Abstract
Close-range first-person field images are important for mobile maize phenotyping because many plant-level traits depend on in-canopy structures that are difficult to ob serve from overhead views. However, post-seedling maize fields create a difficult in stance segmentation setting: stems, leaves, tassels, and neighboring plants are elon gated, repetitive, and strongly occluded. We introduce EgoMaize, a compact benchmark for first-person maize instance segmentation, where the task is to predict ownership consistent plant masks and plant-owned stem/tassel cues from close-range field images with severe same-class overlap. Existing visible-only labels can fragment one physi cal plant into disconnected supervision, while full-amodal labels may require unverifi able completion behind neighboring plants or field objects. EgoMaize therefore uses an evidence-closed annotation workflow for occluded maize regions and assigns unreli able maize regions to ignore rather than background. Baseline results show that pre trained query-based grouping, boundary refinement, and high-resolution crop refine ment help different aspects of the task, but no architecture solves the coupled chal lenges of fine structure recovery, same-class instance ownership, and occlusion reason ing; occlusion-level analysis further shows that performance decreases as plant visi bility becomes more limited. The dataset and code are publicly available at https: //github.com/JaaaaaaaD/EgoMaize.