journal menu
![[Figure 4]](jo5091fig4mag.jpg)
|
|
Figure 4
Primary training of the algorithm on 50 images of target G (G50, panels C and G) or target H (H50, panels D and H), and the evaluation of each of the targets G (left, A–D) or H (right, E–H). Additional training of the algorithm was performed using the data-science-bowl dataset (Caicedo et al., 2019  ) for the segmentation of the cell bodies (panels B and F). The results indicate that specific training on the target crystal is required for optimum performance. Performance indicator values are shown in Table 1. |
![[Figure 4]](jo5091fig4.jpg)
 | JOURNAL OF APPLIED CRYSTALLOGRAPHY |
ISSN: 1600-5767
access
