def main():
parser = argparse.ArgumentParser(
description="Trim Detection weights and save in PyTorch format.")
parser.add_argument(
"--input_dir",
default="",
help="Input directory which contains images and annotations",
type=str,
)
parser.add_argument(
"--output_dir",
default="",
help=
"Output directory which contains images and annotations after data augmentation",
type=str,
)
args = parser.parse_args()
# create the augmentor object
PROBLEM = "instance_segmentation"
ANNOTATION_MODE = "coco"
INPUT_PATH = args.input_dir
GENERATION_MODE = "linear"
OUTPUT_MODE = "coco"
OUTPUT_PATH = args.output_dir + '/'
augmentor = createAugmentor(PROBLEM, ANNOTATION_MODE, OUTPUT_MODE,
GENERATION_MODE, INPUT_PATH,
{"outputPath": OUTPUT_PATH})
# add the augmentation techniques
transformer = transformerGenerator(PROBLEM)
for angle in [0, 90, 180, 270]:
rotate = createTechnique("rotate", {"angle": angle})
augmentor.addTransformer(transformer(rotate))
flip = createTechnique("flip", {"flip": 1})
augmentor.addTransformer(transformer(flip))
crop = createTechnique("crop", {"percentage": 0.7, "startFrom": "TOPLEFT"})
augmentor.addTransformer(transformer(crop))
crop = createTechnique("crop", {
"percentage": 0.7,
"startFrom": "BOTTOMRIGHT"
})
augmentor.addTransformer(transformer(crop))
# apply the augmentation process
augmentor.applyAugmentation()
print('Done!')
def augment_data(train_data_text_file, augmented_path):
#1. copy the images from training dataset
#copy_for_augment(train_data_text_file, raw_path, augmented_path, last_frame)
#2. initliaze augmentor
augmentor, transformer = init_augment(augmented_path)
vFlip = createTechnique("flip", {"flip": 0})
augmentor.addTransformer(transformer(vFlip))
hFlip = createTechnique("flip", {"flip": 1})
augmentor.addTransformer(transformer(hFlip))
hvFlip = createTechnique("flip", {"flip": -1})
augmentor.addTransformer(transformer(hvFlip))
rotate = createTechnique("rotate", {"angle": 90})
augmentor.addTransformer(transformer(rotate))
avgBlur = createTechnique("average_blurring", {"kernel": 5})
augmentor.addTransformer(transformer(avgBlur))
hue = createTechnique("raise_hue", {"power": 0.9})
augmentor.addTransformer(transformer(hue))
total_frames = constants.TOTAL_FRAMES
frames = []
for fname in os.listdir(augmented_path):
if fname.endswith(".jpg"):
frames.append(fname)
for fr in frames:
img, boxes = load_boxes(augmented_path + "/" + fr,
augmented_path + "/" + fr[:-3] + "txt")
ver_img, ver_boxes = aug_operation(vFlip, transformer, img, boxes)
save_augmentation(ver_img, ver_boxes, augmented_path, total_frames)
total_frames += 1
hor_img, hor_boxes = aug_operation(hFlip, transformer, img, boxes)
save_augmentation(hor_img, hor_boxes, augmented_path, total_frames)
total_frames += 1
blur_img, blur_boxes = aug_operation(avgBlur, transformer, img, boxes)
save_augmentation(blur_img, blur_boxes, augmented_path, total_frames)
total_frames += 1
hv_img, hv_boxes = aug_operation(hvFlip, transformer, img, boxes)
save_augmentation(hv_img, hv_boxes, augmented_path, total_frames)
total_frames += 1
hue_img, hue_boxes = aug_operation(hue, transformer, img, boxes)
save_augmentation(hue_img, hue_boxes, augmented_path, total_frames)
total_frames += 1
def apply_eqhisto(input_path, output_path):
PROBLEM = "semantic_segmentation"
ANNOTATION_MODE = "folders"
INPUT_PATH = input_path + "/"
GENERATION_MODE = "linear"
OUTPUT_MODE = "folders"
OUTPUT_PATH= output_path + "/"
LABELS_EXTENSION = ".png"
augmentor = createAugmentor(PROBLEM,ANNOTATION_MODE,OUTPUT_MODE,GENERATION_MODE,INPUT_PATH,{"outputPath":OUTPUT_PATH,"labelsExtension":LABELS_EXTENSION})
transformer = transformerGenerator(PROBLEM)
equalize = createTechnique("equalize_histogram",{})
augmentor.addTransformer(transformer(equalize))
augmentor.applyAugmentation()
print("Equalize histogram results were saved given directory.")
def apply_shearing(input_path, output_path, a = 0.5):
PROBLEM = "semantic_segmentation"
ANNOTATION_MODE = "folders"
INPUT_PATH = input_path + "/"
GENERATION_MODE = "linear"
OUTPUT_MODE = "folders"
OUTPUT_PATH= output_path + "/"
LABELS_EXTENSION = ".png"
augmentor = createAugmentor(PROBLEM,ANNOTATION_MODE,OUTPUT_MODE,GENERATION_MODE,INPUT_PATH,{"outputPath":OUTPUT_PATH,"labelsExtension":LABELS_EXTENSION})
transformer = transformerGenerator(PROBLEM)
rotate = createTechnique("shearing", {"a" : a})
augmentor.addTransformer(transformer(rotate))
augmentor.applyAugmentation()
print("Shearing results were saved given directory.")
def aug_blurring(input_path, output_path, blurr:list):
PROBLEM = "semantic_segmentation"
ANNOTATION_MODE = "folders"
INPUT_PATH = input_path + "/"
GENERATION_MODE = "linear"
OUTPUT_MODE = "folders"
OUTPUT_PATH= output_path + "/"
LABELS_EXTENSION = ".png"
augmentor = createAugmentor(PROBLEM,ANNOTATION_MODE,OUTPUT_MODE,GENERATION_MODE,INPUT_PATH,{"outputPath":OUTPUT_PATH,"labelsExtension":LABELS_EXTENSION})
transformer = transformerGenerator(PROBLEM)
for ker in blurr:
blur = createTechnique("blurring", {"kernel" : ker})
augmentor.addTransformer(transformer(blur))
print("Blurring for kernel = {} is done".format(ker))
augmentor.applyAugmentation()
print("Augmentation results were saved given directory.")
def apply_white_noise(input_path, output_path, sd):
PROBLEM = "semantic_segmentation"
ANNOTATION_MODE = "folders"
INPUT_PATH = input_path + "/"
GENERATION_MODE = "linear"
OUTPUT_MODE = "folders"
OUTPUT_PATH= output_path + "/"
LABELS_EXTENSION = ".png"
augmentor = createAugmentor(PROBLEM,ANNOTATION_MODE,OUTPUT_MODE,GENERATION_MODE,INPUT_PATH,{"outputPath":OUTPUT_PATH,"labelsExtension":LABELS_EXTENSION})
transformer = transformerGenerator(PROBLEM)
for sigma in sd:
white_noise = createTechnique("gaussian_noise", {"mean" : 0,"sigma":sigma})
augmentor.addTransformer(transformer(white_noise))
augmentor.applyAugmentation()
print("White noise results were saved given directory.")
def apply_gamma_correction(input_path, output_path, gammas):
PROBLEM = "semantic_segmentation"
ANNOTATION_MODE = "folders"
INPUT_PATH = input_path + "/"
GENERATION_MODE = "linear"
OUTPUT_MODE = "folders"
OUTPUT_PATH= output_path + "/"
LABELS_EXTENSION = ".png"
augmentor = createAugmentor(PROBLEM,ANNOTATION_MODE,OUTPUT_MODE,GENERATION_MODE,INPUT_PATH,{"outputPath":OUTPUT_PATH,"labelsExtension":LABELS_EXTENSION})
transformer = transformerGenerator(PROBLEM)
for gamma in gammas:
gamma_t = createTechnique("gamma", {"gamma" : gamma})
augmentor.addTransformer(transformer(gamma_t))
augmentor.applyAugmentation()
print("Gamma correction results were saved given directory.")
def apply_dropout(input_path, output_path, percentages):
PROBLEM = "semantic_segmentation"
ANNOTATION_MODE = "folders"
INPUT_PATH = input_path + "/"
GENERATION_MODE = "linear"
OUTPUT_MODE = "folders"
OUTPUT_PATH= output_path + "/"
LABELS_EXTENSION = ".png"
augmentor = createAugmentor(PROBLEM,ANNOTATION_MODE,OUTPUT_MODE,GENERATION_MODE,INPUT_PATH,{"outputPath":OUTPUT_PATH,"labelsExtension":LABELS_EXTENSION})
transformer = transformerGenerator(PROBLEM)
for percentage in percentages:
dropout = createTechnique("dropout", {"percentage" : percentage})
augmentor.addTransformer(transformer(dropout))
augmentor.applyAugmentation()
print("Dropout results were saved given directory.")
img = cv2.imread(image_path)
h, w = img.shape[:2]
lines = [line.strip('\n') for line in open(label_path)]
boxes = []
if lines != ['']:
for line in lines:
words = line.split(' ')
id = words[0]
x = int(float(words[1]) * w - float(words[3]) * w / 2)
y = int(float(words[2]) * h - float(words[4]) * h / 2)
he = int(float(words[4]) * h)
wi = int(float(words[3]) * w)
boxes.append((id, (x, y, wi, he)))
return img, boxes
augmentor = createAugmentor('detection', 'yolo', 'yolo', 'linear', 'dataset', {'outputPath': 'dataset/augmented'})
transformer = transformerGenerator('detection')
v_flip = createTechnique('flip', {'flip': 0})
h_flip = createTechnique('flip', {'flip': 1})
hv_flip = createTechnique('flip', {'flip': -1})
avg_blur_3 = createTechnique('average_blurring', {'kernel': 3})
none = createTechnique('none', {})
augmentor.addTransformer(transformer(v_flip))
augmentor.addTransformer(transformer(h_flip))
augmentor.addTransformer(transformer(hv_flip))
augmentor.addTransformer(transformer(avg_blur_3))
augmentor.addTransformer(transformer(none))
augmentor.applyAugmentation()
# from keras.optimizers import SGD
from clodsa.augmentors.augmentorFactory import createAugmentor
from clodsa.transformers.transformerFactory import transformerGenerator
from clodsa.techniques.techniqueFactory import createTechnique
from clodsa.utils.minivgg import MiniVGGNet
import cv2
import parseConfig
###
config = parseConfig.parseConfig("config_augmentImages.json")
augmentor = createAugmentor(config["problem"], config["annotationMode"],
config["outputMode"], config["generationMode"],
config["inputPath"],
{"outputPath": config["outputPath"]})
transformer = transformerGenerator(config["problem"])
###
# Load the techniques and add them to the augmentor
techniques = [
createTechnique(technique, param)
for (technique, param) in config["augmentationTechniques"]
]
i = 0
for technique in techniques:
augmentor.addTransformer(transformer(technique))
i = i + 1
agm = augmentor.applyAugmentation()
from clodsa.augmentors.augmentorFactory import createAugmentor
from clodsa.transformers.transformerFactory import transformerGenerator
from clodsa.techniques.techniqueFactory import createTechnique
from clodsa.utils.minivgg import MiniVGGNet
import cv2
import parseConfig
config = parseConfig.parseConfig("cats_dogs_folder_folder_linear.json")
augmentor = createAugmentor(config["problem"],config["annotationMode"],config["outputMode"],config["generationMode"],config["inputPath"],
{"outputPath":config["outputPath"]})
transformer = transformerGenerator(config["problem"])
# Load the techniques and add them to the augmentor
techniques = [createTechnique(technique,param) for (technique,param) in config["augmentationTechniques"]]
print("Number of images in input folder")
!ls /home/alex/anaconda3/envs/DD2424-project/DD2424---Project-Covid-19/git/datasets/images | wc -l
img = {}
img["original"] = cv2.imread(config["inputPath"] + "images/cat_1.jpg")
i = 0
plt.figure("original")
# changing to the BGR format of OpenCV to RGB format for matplotlib
plt.imshow(img["original"][:,:,::-1])
for technique in techniques:
augmentor.addTransformer(transformer(technique))
img[config["augmentationTechniques"][i][0]] = technique.apply(img["original"])
PROBLEM = "instance_segmentation"
ANNOTATION_MODE = "coco"
INPUT_PATH = "/Users/sunbowen/Desktop/shellfish_detection/data/val_img"
GENERATION_MODE = "linear"
OUTPUT_MODE = "coco"
OUTPUT_PATH = "/Users/sunbowen/Desktop/shellfish_detection/data/output/"
augmentor = createAugmentor(PROBLEM, ANNOTATION_MODE, OUTPUT_MODE,
GENERATION_MODE, INPUT_PATH,
{"outputPath": OUTPUT_PATH})
transformer = transformerGenerator(PROBLEM)
# Rotations
for angle in [0, 90, 180]:
rotate = createTechnique("rotate", {"angle": angle})
augmentor.addTransformer(transformer(rotate))
# Flip
flip = createTechnique("flip", {"flip": 1})
augmentor.addTransformer(transformer(flip))
# Implementation
augmentor.applyAugmentation()
#
# image_directory = OUTPUT_PATH
# annotation_file = OUTPUT_PATH + 'annotation.json'
#
# example_coco = COCO(annotation_file)
#
# categories = example_coco.loadCats(example_coco.getCatIds())
GENERATION_MODE = "sequential"
OUTPUT_MODE = "coco"
OUTPUT_PATH = "output/"
augmentor = createAugmentor(PROBLEM, ANNOTATION_MODE, OUTPUT_MODE,
GENERATION_MODE, INPUT_PATH,
{"outputPath": OUTPUT_PATH})
transformer = transformerGenerator(PROBLEM)
#for angle in [90,180]:
def sample():
return (50, 50, 256, 256)
t = createTechnique("patchify", {"patchify": sample})
#t = createTechnique("random_object_non_occlusion",
# {"random_images_dir":"/home/fer/repos/CLoDSA/random_instances/",
# "rate":0.5,
# "resize_factor":0.4})
#augmentor.addTransformer(transformer(t))
#t = createTechnique("resize", {"x":20,"y":20})
#augmentor.addTransformer(transformer(t))
#rotate = createTechnique("random_rotate", {"range":[-90,90]})
#augmentor.addTransformer(transformer(rotate))
#flip0 = createTechnique("random_object_occlusion",
# {"random_images_dir":"/home/fer/repos/CLoDSA/random_instances/color"})
