Image Augmentation for Multitask Few-Shot Learning: Agricultural Domain Use-Case

Abstract

Large datasets' availability is catalyzing a rapid expansion of deep learning in general and computer vision in particular. At the same time, in many domains, a sufficient amount of training data is lacking, which may become an obstacle to the practical application of computer vision techniques. This paper challenges small and imbalanced datasets based on the example of a plant phenomics domain. We introduce an image augmentation framework, which enables us to extremely enlarge the number of training samples while providing the data for such tasks as object detection, semantic segmentation, instance segmentation, object counting, image denoising, and classification. We prove that our augmentation method increases model performance when only a few training samples are available. In our experiment, we use the DeepLabV3 model on semantic segmentation tasks with Arabidopsis and Nicotiana tabacum image dataset. The obtained result shows a 9% relative increase in model performance compared to the basic image augmentation techniques.

Plant phenotyping relevance

植物フェノミクス画像を対象に、少数データでのセグメンテーション等を改善する画像拡張フレームワークを開発・評価しており、植物表現型取得・抽出の計算手法が中心である。

Code and data availability

The paper uses the IPPN plant image dataset (a cited prior-work dataset, not a paper-specific deposit) and describes an augmentation framework whose code is only promised for future release ('will be shared as an open source code with the community') with no public URL, repository, or identifier. No paper-specific,公开,

No evidence-backed public reproduction asset is currently recorded.