Figure 2
From: Consumer-grade UAV imagery facilitates semantic segmentation of species-rich savanna tree layers
Visualising the confusion matrix for predictions of DeepLabv3 + trained with 512 × 512 pixel input tiles. Predictions were made on three separate 1-ha test plots spatially distant from the training areas. Model output is depicted on the horizontal axis, and true labels on the vertical axis. Colour intensity corresponds to the number of pixels for the respective combination.
