An Ensemble of Transfer Learning Models for the Prediction of Skin Lesions with Conditional Generative Adversarial Networks

Table 2

Image augmentation techniques.

Flip	50% of horizontal and vertical flip on all images.
Affine	Translation: move each image −20 to +20% per axis; Rotation: rotate each image by −30 to 30 degrees; Scaling: zoom in each image by 0.5 to 1.5 times
Multiply	Multiplication of each image by a random value sampled from [0.8, 1.2].
Linear contrast	Change contrast by equation
	127 + alpha  (v-127)
	V: pixel value
	Alpha: samples from [0.6, 1.4]
Gaussian blur	Blur the images using Gaussian kernel with standard deviation sampled from the interval [0.0, 3.0].