Seed yield prediction in forage plants involves the detection and counting of individual racemes that comprise an inflorescence. However, this task is labor-intensive to perform manually across large numbers of plants and overly complex for classical machine learning techniques due to challenges such as high raceme overlap, large variations in raceme numbers per image and spectral signature similarities between the racemes and the vegetative parts of the plant. To address these challenges, a deep learning-based desktop tool was implemented to count individual racemes in RGB images of Urochloa genotypes, showing different phenological stages and wide variation in number of racemes per plant.