Skip to main content
. 2018 Oct 10;9(11):5353–5367. doi: 10.1364/BOE.9.005353

Fig. 5.

Fig. 5

How the number of training patches affects the performance of the different CNN algorithms. Random weight initialization (Random) is comparable to the Transfer Learning (TFL), Fine-Tuning Layer 5 (FT5), and Fine-Tuning Layer 9 (FT9) methods at 3,000 training patches.