Fig. 3
From: HairNet2: deep learning to quantify cotton leaf hairiness, a complex genetic and environmental trait
Comparison of LTS from human annotations with HairNet2 predictions on the 251 images of the AnnCoT Test split. The scatter plot (A) and the violin plot (B) show that HairNet2 tended to return higher LTS than human annotators. This was not due to the detection of false positives but more likely to the detection by HairNet2 of trichomes which had been missed by human annotators (C, red: human annotations, green: HairNet2 predictions). It is also possible that HairNet2 annotations were slightly wider than human annotations but this could not be quantified