Fig. 5From: DeepCob: precise and high-throughput analysis of maize cob geometry using deep learning with an application in genebank phenomicsImprovement of \(AP@\)[.5:.95] scores during 50 epochs of model updating to different maize cob image datasets (a, b: ImgCross; c, d: ImgDiv). Updating on the COCO initial weights/COCO model (a, c) in comparison to updating on the pre-trained maize model (b, d) depends on different amounts of training images, namely 10, 20, 30, 40 or 50 imagesBack to article page