Fig. 3.

Estimation of fruit number from image analysis and statistical modeling. a Example of inflorescence images, analyzed with ImageJ to extract individual skeleton descriptors after segmentation and 2D skeletonization. b Manual counting (purple dots) of fruit number on a subset of inflorescence images. c Series of cross-validation performed for different predictive models with different training population size (x axis). Dots represent mean prediction accuracy, measured as Pearson’s coefficient of correlation (r²) between observed and predicted values. Error bars represent 95% CI across 100 random permutations of the training dataset. d Correlation between observed and predicted values for cross-validation of the best model obtained with stepwise regression, performed 60 individuals to train the model, and tested on 100 individuals not used to train the model