Table 2 Dataset statistics and distribution.

From: Novel dual gland GAN architecture improves human protein localization classification using salivary and pituitary gland inspired loss functions

Parameter

Value

Number of cell types

17

Number of localization classes

19

Image resolution range

1728 × 1728 to 3072 × 3072 pixels

Image formats

PNG (8-bit), TIF (16-bit)

Fluorescent channels

4 (green, blue, red, yellow)

Total training images

42,774

Total test images

11,702

Average cells per image

8.7

Most common localization class

Nucleoplasm (27.4%)

Least common localization class

Mitotic spindle (1.3%)

Class imbalance ratio

21:1 (most to least common)