Fig. 6From: DeepCob: precise and high-throughput analysis of maize cob geometry using deep learning with an application in genebank phenomicsDetection of cob and ruler after model updating the pretained maize model with different image datasets. a Updating with 10 training images from ImgCross. The original maize model detected only one cob (epoch 0). After one epoch of model updating both cobs were accurately segmented and after epoch 12 the different ruler element was detected. Photo credit: K. Schmid, University of Hohenheim. b Segmentation of various genebank images after updating for 25 epochs with 20 training images from ImgDiv. Photo credits: https://nexusmedianews.com/drought-is-crippling-small-farmers-in-mexico-with-consequences-for-everyone-else-photos-73b35a01e4dd (Left) https://www.ars.usda.gov/ARSUserFiles/50301000/Races_of_Maize/RoM_Paraguay_0_Book.pdf (Center) Right: CIMMYT, https://flic.kr/p/9h9X6B. All photos are available under a Creative Commons License. c Segmentation of cobs and rulers in post-harvest images of the Swiss Rheintaler Ribelmais landrace with the best model from ImgCross without updating on these images. Photo credit: Benedikt Kogler, Verein Rheintaler Ribelmais e.V., SwitzerlandBack to article page