posted on 2022-01-19, 06:40authored byVivien Rolland, Moshiur R. Farazi, Warren C. Conaty, Deon Cameron, Shiming Liu, Lars Petersson, Warwick N. Stiller
Additional file 1: Table S1. Average [min, max] accuracy of HairNet model on leaf-based splits with training dataset size, as reported in Fig. 8a. Table S2. Average [min, max] accuracy of HairNet model on year-based splits with training dataset size, as reported in Fig. 8b. Table S3. Average [min, max] accuracy of HairNet model on environment-based splits with training dataset size, as reported in Fig. 8c. Figure S1. Effect of varying the probability (p) of Random Vertical (RV) and Random Horizontal (RH) flip data augmentation on model accuracy (Image Accuracy [IA]). Figure S2. Effect of different weight initialisation methods in the classification neural network on model accuracy (Image Accuracy [IA]). Figure S3. Prediction of HairNet model on random examples from the whole dataset. Figure S4. Prediction of HairNet model on random first image (most proximal) examples from the whole dataset. Figure S5. Effect of Gaussian noise on model accuracy. Figure S6. Normalized confusion matrices of HairNet predictions on leaf-based splits. Figure S7. Normalized confusion matrices of HairNet predictions on year-based splits. Figure S8. Normalized confusion matrices of HairNet predictions on environment-based splits.
Funding
Commonwealth Scientific and Industrial Research Organisation