image-augmentation-for-landmarks

python class for image augmentation with landmarks or keypoints Data in mind

the code design and the functions are similar to waht there in keras API image_Data_generator most of the functionality and more are coming soon with other improvements

How to use it

clone the repo then place the augtool.py file in the project folder

1. first initialize the class

gen = image_generator_landmarksAware(image=img,
                                     keypoints=points,
                                     rotate_range=(-50,70),
                                     shift_range=0.12,
                                     brightness_range=(0.4,1.3),   
                                     noise_range=0.04,
                                     allow_repeate=True,
                                     blur_range=(5,3),
                                     horizontal_flip=True,
                                     zoom_range=(0.8,1.4),
                                     target_shape=(256,256,3),
                                     sharpen=False,
                                     epochs=3

the default values for those are 0.0 None or False so feel free to leave them empty and just type what you want to augment

2.generate 20 augmented images on the Fly

for image, kpoints in gen.generate(20):
    display_with_landmark(image, kpoints)

the display_with_landmark should be in the augtool so simple from augtools import * or augtools.display_with_landmark should be enough.

what we just did is generating 20 images and plot them with corresponding landmarks for testing

3.save the augmented images to the directory

as you can see we need to save the images to our dirctory with these augmentation simple python scripting like these can do it for you

for image, k in gen.generate(20):
    image_name = f'augmented_{uuid.uuid1()}.jpeg' # get random image name
    plt.imsave(f'augmented_images/{image_name}, img) # save the augmented image
    record = {"image_name" : image_name,"coordinates" : lms.tolist()}
    data_list.append(record)

with open('../data/augkeypointsdata.json', 'w') as f:
    json.dump(data_list, f)

what will be next

improving the code design more
more Documentation and examples
support for list of images as input instead of one at a time
ability to generate images as batches
add more image manipulation functions