From Semantic To Instance: A Semi-Self-Supervised Learning Approach

TL;DR

This paper introduces a semi-self-supervised learning method for instance segmentation that minimizes manual annotation, focusing on shape and texture, and achieves state-of-the-art results in agricultural and general datasets.

Contribution

The paper presents GLMask, a novel image-mask representation and pipeline that significantly improves instance segmentation performance with minimal manual labels.

Findings

Achieved 98.5% mAP@50 on wheat head segmentation

Improved COCO dataset performance by over 12.6% mAP@50

Outperformed conventional models in dense, occluded object scenarios

Abstract

Instance segmentation is essential for applications such as automated monitoring of plant health, growth, and yield. However, extensive effort is required to create large-scale datasets with pixel-level annotations of each object instance for developing instance segmentation models that restrict the use of deep learning in these areas. This challenge is more significant in images with densely packed, self-occluded objects, which are common in agriculture. To address this challenge, we propose a semi-self-supervised learning approach that requires minimal manual annotation to develop a high-performing instance segmentation model. We design GLMask, an image-mask representation for the model to focus on shape, texture, and pattern while minimizing its dependence on color features. We develop a pipeline to generate semantic segmentation and then transform it into instance-level segmentation. The proposed approach substantially outperforms the conventional instance segmentation models, establishing a state-of-the-art wheat head instance segmentation model with mAP@50 of 98.5%. Additionally, we assessed the proposed methodology on the general-purpose Microsoft COCO dataset, achieving a significant performance improvement of over 12.6% mAP@50. This highlights that the utility of our proposed approach extends beyond precision agriculture and applies to other domains, specifically those with similar data characteristics.