Data augmentation with grid search idea. You can generate as many images as you want according to different data augmentation combinations based on grid search.

grid_data_augmentation.py

'''
-----------------------------------------------
File Name: grid search$
Description:
Author: Jing$
Date: 8/23/2021$
-----------------------------------------------
'''

import numpy as np
import copy
import glob, os
import cv2
import SimpleITK as sitk ## using simpleITK to load and save medical data.

class Solution(object):
    result = []
    temp = []

    def subsets(self, nums):
        """
        :type nums: List[int]
        :rtype: List[List[int]]
        """
        del self.result[:]
        del self.temp[:]

        if nums == []:
            return self.result
        else:
            self.backtrack(nums, 0)
            return self.result

    def backtrack(self, nums, startIndex):

        length = len(nums)
        self.result.append(copy.deepcopy(self.temp))  # collect all the nodes
        if startIndex >= length:
            return

        for i in range(startIndex, length):  # startIndex determins travel width, ordered
            self.temp.append(nums[i])
            self.backtrack(nums, i + 1)  # i determins travel depth, i+1 means no duplication
            self.temp = self.temp[:-1]

def aug_rotate(image):
    angle = 10
    height, width = image.shape
    rotate_around = (width // 2, height // 2)
    M = cv2.getRotationMatrix2D(rotate_around, angle, 1)
    image = cv2.warpAffine(image, M, (width, height))
    return image
def aug_rotate_r(image):
    angle = -5
    height, width = image.shape
    rotate_around = (width // 2, height // 2)
    M = cv2.getRotationMatrix2D(rotate_around, angle, 1)
    image = cv2.warpAffine(image, M, (width, height))
    return image
def aug_trans_x(image):
    shiftX = 5
    shiftY = 0
    M = np.float32([ [1, 0, shiftX], [0, 1, shiftY] ])
    height, width = image.shape
    image = cv2.warpAffine(image, M, (width, height))
    return image
def aug_trans_y(image):
    shiftX = 0
    shiftY = 5
    M = np.float32([ [1, 0, shiftX], [0, 1, shiftY] ])
    height, width = image.shape
    image = cv2.warpAffine(image, M, (width, height))
    return image
def aug_flip_h(image):
    image = cv2.flip(image, 1)  # 1 for Horizontal,0 for vertical
    return image
def aug_flip_v(image):
    image = cv2.flip(image, 0)  # 1 for Horizontal,0 for vertical
    return image
def aug_shear_x(image):
    shearX = 0.1
    shearY = 0
    shearM = np.array([
        [1, shearX, 0],# x direction rate is tan =0.1
        [shearY, 1, 0]
    ])
    height, width = image.shape
    img_shear = cv2.warpAffine(image, shearM,(width, height))
    return img_shear
def aug_shear_y(image):
    shearX = 0
    shearY = 0.1
    shearM = np.array([
        [1, shearX, 0],# x direction rate is tan =0.3
        [shearY, 1, 0]
    ])
    height, width = image.shape
    img_shear = cv2.warpAffine(image, shearM,(width, height))
    return img_shear
def aug_gauss(image):
    blur = cv2.GaussianBlur(image, (3,3), 0)#kernel size, sigma, bigger, blurer
    return blur
def aug_gamma_correct(image):
    g = 1.5
    out = np.power(image / float(np.max(image)), 1 / g)# gamma bigger, brighter
    out = out*255.0
    out = out.astype(np.uint8)
    return out

#image = cv2.imread('./example/p001_fm01.png')
# example
#image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
# imager = aug_gamma_correct(image)
# cv2.imshow('result',imager)
# cv2.waitKey()

func_list=['aug_rotate','aug_rotate_r','aug_flip_h','aug_flip_v','aug_trans_x','aug_trans_y',
           'aug_shear_x','aug_shear_y','aug_gauss','aug_gamma_correct']
func_index=[0,1,2,3,4,5,6,7,8,9]

solu = Solution()
aug_array = sorted(solu.subsets(func_index),key=len)

# test
print(aug_array[:29]) # subsets
print(len(aug_array[:29])) # the length of selected subsets
image_path = './example/'
def grid_aug_image():
    for i in range(1,len(aug_array[:29])):
        if len(aug_array[i])==1:
            print(aug_array[i][0])
            imagelist = sorted(glob.glob(os.path.join(image_path, '*.png')))  # sorted as name
            for m in range(len(imagelist)):
                print(imagelist[m])
                image = cv2.imread(imagelist[m])
                image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
                image_aug = globals()[func_list[aug_array[i][0]]](image)
                cv2.imshow('result',image_aug)
                path = "./"+str(aug_array[i][0])+'/'
                if not os.path.exists(path): os.makedirs(path)
                cv2.imwrite(path + imagelist[m][-13:-3] + '_'+str(aug_array[i][0])  + imagelist[m][-4:], image_aug)

        if len(aug_array[i]) == 2:
            print(aug_array[i])
            imagelist = sorted(glob.glob(os.path.join(image_path, '*.png')))
            for m in range(len(imagelist)):
                print(imagelist[m])
                image = cv2.imread(imagelist[m])
                image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
                image = globals()[func_list[aug_array[i][0]]](image)
                image = globals()[func_list[aug_array[i][1]]](image)
                path = "./"+str(aug_array[i][0])+str(aug_array[i][1])+'/'
                if not os.path.exists(path): os.makedirs(path)
                cv2.imwrite(path+imagelist[m][-13:-4]+'_'+str(aug_array[i][0])+str(aug_array[i][1])+imagelist[m][-4:],image)

img_path = 'dataset/img_ori/'
save_path = 'dataset/'
def grid_aug_medimage():
    for i in range(1, len(aug_array[:29])):
        if len(aug_array[i]) == 1:
            print(aug_array[i][0])
            imagelist = sorted(glob.glob(os.path.join(img_path, '*.nii.gz')))
            for m in range(len(imagelist)):
                print(imagelist[m])
                itk_img = sitk.ReadImage(imagelist[m])
                image = sitk.GetArrayFromImage(itk_img)
                print(image.shape[0])
                for k in range(0, image.shape[0]):  # for each slice
                    image[k] = globals()[func_list[aug_array[i][0]]](image[k])

                image_aug = sitk.GetImageFromArray(image)
                path = save_path + str(aug_array[i][0]) + '/'
                if not os.path.exists(path): os.makedirs(path)
                sitk.WriteImage(image_aug, path + imagelist[m][-16:-7] + '_' + str(aug_array[i][0]) + imagelist[m][-7:])

        if len(aug_array[i]) == 2:
            print(aug_array[i])
            imagelist = sorted(glob.glob(os.path.join(img_path, '*.nii.gz')))
            for m in range(len(imagelist)):
                print(imagelist[m])
                itk_img = sitk.ReadImage(imagelist[m])
                image = sitk.GetArrayFromImage(itk_img)

                for k in range(0, image.shape[0]):  # for each slice
                    image[k] = globals()[func_list[aug_array[i][0]]](image[k]) # call the augmentation functions in the strings
                    image[k] = globals()[func_list[aug_array[i][1]]](image[k])

                image_aug = sitk.GetImageFromArray(image)
                path = save_path + str(aug_array[i][0]) + str(aug_array[i][1]) + '/'
                if not os.path.exists(path): os.makedirs(path) # create new folder and save augmented images
                sitk.WriteImage(image_aug, path + imagelist[m][-16:-7] + '_' + str(aug_array[i][0]) + str(aug_array[i][1]) + imagelist[m][-7:])

#grid_aug_medimage()

def Info(images_path):
    imagelist = sorted(glob.glob(os.path.join(images_path, '*.gz')))  
    for i in (range(len(imagelist))):
        print(imagelist[i][-19:]) # print the filename

Info('dataset9_aug/23/')

For example, if you have 200 images, you want to augment 20000 new different images, which is 100 times of original images. In other words, you need to augment the images for 100 times.
In this code, there are 10 kinds of single data augmentation methods. When you use the grid search idea, there are totally 1024 subsets (data augmentation combinations). You just use 100 subsets of them so that can generate 20000 images without duplication.