Skip to main content
. 2023 Jan 9;12:1044026. doi: 10.3389/fonc.2022.1044026

Table 1.

Transformations available in Semi-His-Net.

Transformation Description Parameter Range
Rotate Rotates the picture by degrees. θ [-20, 20]
Fliplr Flips the picture horizontally with a probability of Λ. Λ [0, 1]
Flipud Flips the picture vertically with a probability of Λ. Λ [0, 1]
Brightness Converts the picture to a colorspace with a brightness-related channel, adds between −α and α, then converts back to the input colorspace (RGB). α [-30, 30]
Contrast Adjusts contrast by scaling each pixel to
127 + α*(v−127).
α [0.1, 1.9]
Cutout Fills one or more random rectangular patches in the picture using a fill mode. n [1, 5]
Dropout Sets p percent of pixels in picture to zero. p [0, 0.1]
Sharpness Sharpens the picture and covers the results with the initial picture with a strength of Λ. Λ [0, 1]
Shear x Shears the picture alongside the horizontal axis in rate S. S [-0.3, 0.3]
Shear y Shears the picture alongside the vertical axis in rate S. S [-0.3, 0.3]
Change Colorspace Converts to HSV colorspace and adds a value between 0 and α to Hue channel, then converts back to the input colorspace (RGB). α [0, 50]
Solarize Thresholds all pixels over value of T. T [0, 256]
Translate x Transforms the picture horizontally by (Λ × image width) pixels. Λ [-0.3, 0.3]
Translate y Transforms the picture vertically by (Λ × image high) pixels. Λ [-0.3, 0.3]
Gaussian Noise Adds gaussian noise to the picture, sampled normal distribution on per pixel N(0,s). s [0, 0.2]
Gaussian Blur Blurs picture using a gaussian kernel σ. σ [0, 3]
Color Temperature Changes the temperature to Kelvin value between α and β. α and β [1000, 40000]
Invert Inverts the image and replaces the pixels of the original image with a ratio of Λ. Λ [0, 1]