def sroie_refine():
aug_list = []
stage_0, stage_1, stage_2, stage_3 = 1536, 2048, 768, 512
# Pad the height to stage_0
aug_list.append(
augmenters.PadToFixedSize(width=1, height=stage_0, pad_cval=255))
# Resize its height to stage_1, note that stage_0 is smaller than stage_1
# so that the font size could be increased for most cases.
aug_list.append(
augmenters.Resize(size={
"height": stage_1,
"width": "keep-aspect-ratio"
}))
# Crop a stage_2 x stage_2 area
aug_list.append(augmenters.CropToFixedSize(width=stage_2, height=stage_2))
# In case the width is not enough, pad it to stage_2 x stage_2
aug_list.append(
augmenters.PadToFixedSize(width=stage_2, height=stage_2, pad_cval=255))
# Resize to stage_3 x stage_3
aug_list.append(
augmenters.Resize(size={
"height": stage_3,
"width": stage_3
}))
# Perform Flip
aug_list.append(augmenters.Fliplr(0.33, name="horizontal_flip"))
aug_list.append(augmenters.Flipud(0.33, name="vertical_flip"))
return aug_list
def __init__(self):
self.all_hand = []
for suffix in ['.jpg', '.png', '.JPG', '.PNG']:
for f in ['hand', 'egohands_extracted', 'GTEA_extracted']:
self.all_hand += glob.glob(
os.path.join(cfg.path.data_dir, f, '*' + suffix))
self.all_shape = []
for suffix in ['.jpg', '.png', '.JPG', '.PNG']:
for f in ['ShapePick']:
self.all_shape += glob.glob(
os.path.join(cfg.path.data_dir, f, '*' + suffix))
self.hand_augmentor = iaa.Sequential([
iaa.PadToFixedSize(cfg.imsize, cfg.imsize),
iaa.Fliplr(0.5),
iaa.Affine(rotate=(-180, 180),
scale=(0.4, 0.6),
translate_percent={
'x': (-0.5, 0.5),
'y': (-0.5, 0.5)
})
])
self.shape_augmentor = iaa.Sequential([
iaa.PadToFixedSize(cfg.imsize, cfg.imsize),
iaa.Fliplr(0.5),
iaa.Affine(rotate=(-180, 180),
scale=(0.6, 0.8),
translate_percent={
'x': (-0.5, 0.5),
'y': (-0.5, 0.5)
})
])
def aug_sroie(args, bg_color=255):
aug_list = []
stage_0, stage_1, stage_2, stage_3 = 1536, 2048, 768, 512
# Pad the height to stage_0
aug_list.append(
augmenters.PadToFixedSize(width=1, height=stage_0, pad_cval=bg_color))
# Resize its height to stage_1, note that stage_0 is smaller than stage_1
# so that the font size could be increased for most cases.
aug_list.append(
augmenters.Resize(size={
"height": stage_1,
"width": "keep-aspect-ratio"
}))
# increase the aspect ratio
aug_list.append(
augmenters.Sometimes(
args.augment_zoom_probability,
augmenters.Affine(scale=(args.augment_zoom_lower_bound,
args.augment_zoom_higher_bound))))
# Crop a stage_2 x stage_2 area
aug_list.append(augmenters.CropToFixedSize(width=stage_2, height=stage_2))
# In case the width is not enough, pad it to stage_2 x stage_2
aug_list.append(
augmenters.PadToFixedSize(width=stage_2,
height=stage_2,
pad_cval=bg_color))
# Resize to stage_3 x stage_3
#aug_list.append(augmenters.Resize(size={"height": stage_3, "width": stage_3}))
# Perform Flip
aug_list.append(augmenters.Fliplr(0.33, name="horizontal_flip"))
aug_list.append(augmenters.Flipud(0.33, name="vertical_flip"))
return aug_list
def __call__(self, img, anno):
labels = anno[0].numpy()
gts = anno[1].numpy()
np_img = np.asarray(img)
# get BoundingBoxesOnImage
bbs = []
for gt in gts:
bbs.append(BoundingBox(x1=gt[0], y1=gt[1], x2=gt[2], y2=gt[3]))
bbs_on_img = BoundingBoxesOnImage(bbs, shape=np_img.shape)
# draw_img = bbs_on_img.draw_on_image(np_img, size=2)
height = np_img.shape[0]
width = np_img.shape[1]
if height >= width:
length = height
else:
length = width
# position is must, because the bbs and images are augmented separately
self.seq = iaa.Sequential([
iaa.PadToFixedSize(width=length, height=length, position=self.position)
])
# apply augment
image_aug = self.seq.augment_image(np_img)
bbs_aug = self.seq.augment_bounding_boxes(bbs_on_img).bounding_boxes
gts = []
for bb in bbs_aug:
gts.append([bb.x1, bb.y1, bb.x2, bb.y2])
gts = torch.from_numpy(np.array(gts))
return image_aug, (anno[0], gts)
def overlay(args):
'''
Build a dataset augmenting logos images found under a base_dir
to be overlayed on a random bg image
'''
logos = list(
get_labeled_images(
pathlib.Path(args.base_dir).resolve(), args.label_filter))
bg_paths = get_file_paths(args.background_dir)
out_size = args.out_size
# convert the image to a set size
base = [
iaa.Resize({
"shorter-side": 256,
"longer-side": "keep-aspect-ratio"
})
]
augmenters = build_augmenters(args, augmenters=base)
augmenters.append(iaa.PadToFixedSize(width=out_size[0],
height=out_size[1]))
overlay_aug = iaa.Sequential(augmenters)
data = overlay_images_backgrounds(overlay_aug, logos, bg_paths, out_size,
args.count)
process_command(args, logos, data)
def augment_3(self, image, depth, lidar, crop_size, degree):
# rsz = iaa.Resize({"height": resize_size[0], "width": resize_size[1]})
seq = iaa.Sequential(
[
iaa.PadToFixedSize(height=crop_size[0],
width=crop_size[1]), # 保证可crop
iaa.CropToFixedSize(height=crop_size[0],
width=crop_size[1]), # random crop
iaa.Fliplr(0.5),
# iaa.Flipud(0.5),
iaa.Rotate((-degree, degree)),
iaa.GammaContrast((0.9, 1.1)),
iaa.Multiply((0.9, 1.1)),
],
random_order=True)
depth, lidar = np.expand_dims(depth, 2), np.expand_dims(lidar, 2)
tmp = np.concatenate((depth, lidar), axis=2)
tmp = (tmp * 1000).astype(np.int32) # 米单位*1000保留精度
tmp = SegmentationMapsOnImage(tmp, shape=tmp.shape)
# image, tmp = rsz(image=image, segmentation_maps=tmp)
image, tmp = seq(image=image, segmentation_maps=tmp)
tmp = tmp.arr
tmp = tmp.astype(np.float32) / 1000 # 再转回米
depth, lidar = tmp[:, :, 0], tmp[:, :, 1]
return image, depth, lidar
def chapter_augmenters_padtofixedsize():
fn_start = "size/padtofixedsize"
aug_cls = iaa.PadToFixedSize
aug = aug_cls(width=100, height=100)
run_and_save_augseq(fn_start + ".jpg",
aug, [ia.quokka(size=(80, 80)) for _ in range(4 * 2)],
cols=4,
rows=2)
aug = aug_cls(width=100, height=100, position="center")
run_and_save_augseq(fn_start + "_center.jpg",
aug, [ia.quokka(size=(80, 80)) for _ in range(4 * 1)],
cols=4,
rows=1)
aug = aug_cls(width=100, height=100, pad_mode=ia.ALL)
run_and_save_augseq(fn_start + "_pad_mode.jpg",
aug, [ia.quokka(size=(80, 80)) for _ in range(4 * 2)],
cols=4,
rows=2)
aug = iaa.Sequential([
iaa.PadToFixedSize(width=100, height=100),
iaa.CropToFixedSize(width=100, height=100)
])
run_and_save_augseq(fn_start + "_with_croptofixedsize.jpg",
aug, [ia.quokka(size=(80, 120)) for _ in range(4 * 2)],
cols=4,
rows=2)
def __getitem__(self, idx):
img, seg = self.get_data(idx)
img = img[..., ::-1]
h, w, c = img.shape
if self.rect:
scale = min(self.img_size[0] / w, self.img_size[1] / h)
resize = iaa.Sequential([
iaa.Resize({
'width': int(w * scale),
'height': int(h * scale)
}),
iaa.PadToFixedSize(*self.img_size,
pad_cval=[123.675, 116.28, 103.53],
position='center')
])
else:
resize = iaa.Resize({
'width': self.img_size[0],
'height': self.img_size[1]
})
img = resize.augment_image(img)
seg = resize.augment_segmentation_maps(seg)
# augment
if self.augments is not None:
augments = self.augments.to_deterministic()
img = augments.augment_image(img)
seg = augments.augment_segmentation_maps(seg)
img = img.transpose(2, 0, 1)
img = np.ascontiguousarray(img)
seg = seg.get_arr()
return torch.ByteTensor(img), torch.ByteTensor(seg)
def test_sequential(
sample_image, sample_keypoints, width, height,
):
# Guarantee that images smaller than crop size are handled fine
very_small_image = sample_image[:50, :50]
aug = iaa.Sequential(
[
iaa.PadToFixedSize(width, height),
augmentation.KeypointAwareCropToFixedSize(width, height),
]
)
images_aug, keypoints_aug = aug(
images=[very_small_image], keypoints=[sample_keypoints],
)
assert len(images_aug) == len(keypoints_aug) == 1
assert all(im.shape[:2] == (height, width) for im in images_aug)
# Ensure at least a keypoint is visible in each crop
assert all(len(kpts) for kpts in keypoints_aug)
# Test passing in a batch of frames
n_samples = 8
images_aug, keypoints_aug = aug(
images=[very_small_image] * n_samples, keypoints=[sample_keypoints] * n_samples,
)
assert len(images_aug) == len(keypoints_aug) == n_samples
def get_item(self, idx):
img = cv2.imread(self.data[idx][0])
img = img[:, :, ::-1]
h, w, c = img.shape
if self.rect:
scale = min(self.img_size[0] / w, self.img_size[1] / h)
resize = iaa.Sequential([
iaa.Resize({
'width': int(w * scale),
'height': int(h * scale)
}),
iaa.PadToFixedSize(*self.img_size, position='center')
])
else:
resize = iaa.Resize({
'width': self.img_size[0],
'height': self.img_size[1]
})
img = resize.augment_image(img)
# augment
if self.augments is not None:
augments = self.augments.to_deterministic()
img = augments.augment_image(img)
img = img.transpose(2, 0, 1)
img = np.ascontiguousarray(img)
return torch.ByteTensor(img), self.data[idx][1]
def inference(model, img, img_size=(64, 64), rect=False):
img = img[:, :, ::-1]
h, w, c = img.shape
if rect:
scale = min(img_size[0] / w, img_size[1] / h)
resize = ia.Sequential([
ia.Resize({
'width': int(w * scale),
'height': int(h * scale)
}),
ia.PadToFixedSize(*img_size,
position='center')
])
else:
resize = ia.Resize({'width': img_size[0], 'height': img_size[1]})
img = resize.augment_image(img)
img = img.transpose(2, 0, 1)
img = np.ascontiguousarray(img)
imgs = torch.FloatTensor([img]).to(device)
imgs -= torch.FloatTensor([123.675, 116.28,
103.53]).reshape(1, 3, 1, 1).to(imgs.device)
imgs /= torch.FloatTensor([58.395, 57.12,
57.375]).reshape(1, 3, 1, 1).to(imgs.device)
preds = model(imgs)[0].softmax(0).cpu().numpy()
return preds
def YOLO():
"""
Data augmentation model for YOLOv3 training
"""
return iaa.Sequential([
iaa.KeepSizeByResize(
iaa.Affine(
scale=iap.Normal(1, 0.125),
translate_percent=0.1,
cval=128,
)),
iaa.Fliplr(0.5),
iaa.Resize({
"height": iap.Normal(1, 0.1),
"width": iap.Normal(1, 0.1)
}),
iaa.Resize({
"longer-side": 416,
"shorter-side": "keep-aspect-ratio"
}),
iaa.PadToFixedSize(416, 416, pad_cval=128),
iaa.MultiplyHueAndSaturation(mul_hue=iap.Uniform(0, 2),
mul_saturation=iap.Uniform(1 / 1.5, 1.5)),
iaa.AssertShape((None, 416, 416, 3)),
])
def __init__(self, size):
self.seq = iaa.Sequential(
[
iaa.Resize((0.5, 2.0)), # random scale
iaa.PadToFixedSize(width=size[0], height=size[1]),
iaa.CropToFixedSize(width=size[0], height=size[1]),
],
random_order=False)
def __getitem__(self, idx):
img, kps = self.get_data(idx)
img = img[..., ::-1]
h, w, c = img.shape
if self.rect:
scale = min(self.img_size[0] / w, self.img_size[1] / h)
resize = ia.Sequential([
ia.Resize({
'width': int(w * scale),
'height': int(h * scale)
}),
ia.PadToFixedSize(*self.img_size,
pad_cval=[123.675, 116.28, 103.53],
position='center')
])
else:
resize = ia.Resize({
'width': self.img_size[0],
'height': self.img_size[1]
})
img = resize.augment_image(img)
kps = resize.augment_polygons(kps)
# augment
if self.augments is not None:
augments = self.augments.to_deterministic()
img = augments.augment_image(img)
kps = augments.augment_polygons(kps)
heats = [np.zeros(img.shape[:2])] * len(self.classes)
for kp in kps.polygons:
c = kp.label
point = kp.exterior.astype(np.int32)
x = np.arange(img.shape[1], dtype=np.float)
y = np.arange(img.shape[0], dtype=np.float)
xx, yy = np.meshgrid(x, y)
# evaluate kernels at grid points
xxyy = np.c_[xx.ravel(), yy.ravel()]
sigma = 10 # 65.9 # math.sqrt(- math.pow(100, 2) / math.log(0.1))
xxyy -= point
x_term = xxyy[:, 0]**2
y_term = xxyy[:, 1]**2
exp_value = -(x_term + y_term) / 2 / pow(sigma, 2)
zz = np.exp(exp_value)
heat = zz.reshape(img.shape[:2])
heats[c] = heat
# cv2.imshow('c', (heat * 255).astype(np.uint8))
# cv2.waitKey(0)
heats = np.stack(heats, 0)
img = img.transpose(2, 0, 1)
img = np.ascontiguousarray(img)
return torch.ByteTensor(img), torch.FloatTensor(heats)
def resize_with_pad(images, image_size=224):
return iaa.Sequential([
iaa.Resize({
"longer-side": image_size,
"shorter-side": "keep-aspect-ratio"
}),
iaa.PadToFixedSize(width=image_size,
height=image_size,
pad_mode=["constant", "edge", "median"])
])(images=images)
def grid(images, rows, cols, border=1, border_color=255):
nb_images = len(images)
cell_height = max([image.shape[0] for image in images])
cell_width = max([image.shape[1] for image in images])
channels = set([image.shape[2] for image in images])
assert len(channels) == 1
nb_channels = list(channels)[0]
if rows is None and cols is None:
rows = cols = int(math.ceil(math.sqrt(nb_images)))
elif rows is not None:
cols = int(math.ceil(nb_images / rows))
elif cols is not None:
rows = int(math.ceil(nb_images / cols))
assert rows * cols >= nb_images
cell_height = cell_height + 2 * border
cell_width = cell_width + 2 * border
width = cell_width * cols
height = cell_height * rows
grid = np.zeros((height, width, nb_channels), dtype=np.uint8)
cell_idx = 0
for row_idx in range(rows):
for col_idx in range(cols):
if cell_idx < nb_images:
image = images[cell_idx]
border_top = border_right = border_bottom = border_left = border
#if row_idx > 1:
#border_top = 0
#if col_idx > 1:
#border_left = 0
#image = np.pad(image, ((border_top, border_bottom), (border_left, border_right), (0, 0)), mode="constant", constant_values=border_color)
image = iaa.pad(image,
top=border_top,
right=border_right,
bottom=border_bottom,
left=border_left,
mode="constant",
cval=border_color)
image = iaa.PadToFixedSize(height=cell_height,
width=cell_width,
position="center")(image=image)
cell_y1 = cell_height * row_idx
cell_y2 = cell_y1 + image.shape[0]
cell_x1 = cell_width * col_idx
cell_x2 = cell_x1 + image.shape[1]
grid[cell_y1:cell_y2, cell_x1:cell_x2, :] = image
cell_idx += 1
#grid = np.pad(grid, ((border, 0), (border, 0), (0, 0)), mode="constant", constant_values=border_color)
return grid
def __init__(self, src_dir, is_train=True, target_dim=(256, 256)):
self.target_dim = target_dim
self.object_info = dict()
self.class_weights = []
self.image_filenames = []
self.annotation_filenames = []
self.is_train = is_train
if is_train:
images_dir = src_dir + "images/training/"
annotations_dir = src_dir + "annotations/training/"
if target_dim is None: # used in case of class weight computation
self.transform = transforms.Compose([ToTensor()])
else:
self.size_aug = iaa.Sequential([
iaa.PadToFixedSize(target_dim[0], target_dim[1]),
iaa.CropToFixedSize(target_dim[0], target_dim[1])
])
self.property_aug = iaa.Sequential([
iaa.Multiply((0.8, 1.2), per_channel=0.2),
iaa.AdditiveGaussianNoise(loc=0,
scale=(0.0, 0.01 * 255),
per_channel=0.3),
iaa.AddToHueAndSaturation((-20, 20))
])
else:
images_dir = src_dir + "images/validation/"
annotations_dir = src_dir + "annotations/validation/"
self.size_aug = iaa.Sequential([
iaa.PadToFixedSize(target_dim[0], target_dim[1]),
iaa.CropToFixedSize(target_dim[0], target_dim[1])
])
files = os.listdir(images_dir)
for file in files:
filename = file.split(".")[0]
self.image_filenames.append(images_dir + filename + ".jpg")
self.annotation_filenames.append(annotations_dir + filename +
".png")
print(len(self.image_filenames))
def __init__(self):
self.all_celeba = []
for suffix in ['.jpg', '.png', '.JPG', '.PNG']:
self.all_celeba += glob.glob(os.path.join(CELEBA_DIR,
'*' + suffix))
self.all_hand = []
for suffix in ['.jpg', '.png', '.JPG', '.PNG']:
for f in ['hand', 'egohands_extracted', 'GTEA_extracted']:
self.all_hand += glob.glob(
os.path.join(OCCLUSION_DIR, f, '*' + suffix))
self.all_shape = []
for suffix in ['.jpg', '.png', '.JPG', '.PNG']:
for f in ['ShapePick']:
self.all_shape += glob.glob(
os.path.join(OCCLUSION_DIR, f, '*' + suffix))
self.hand_augmentor = iaa.Sequential([
iaa.PadToFixedSize(OCCLUSION_WIDTH, OCCLUSION_HEIGHT),
iaa.Fliplr(0.5),
iaa.Affine(rotate=(-180, 180),
scale=(0.4, 0.6),
translate_percent={
'x': (-0.5, 0.5),
'y': (-0.45, 0.45)
})
])
self.shape_augmentor = iaa.Sequential([
iaa.PadToFixedSize(OCCLUSION_WIDTH, OCCLUSION_HEIGHT),
iaa.Fliplr(0.5),
iaa.Affine(rotate=(-180, 180),
scale=(0.6, 0.8),
translate_percent={
'x': (-0.45, 0.45),
'y': (-0.45, 0.45)
})
])
def augment_image(image, bbs, image_size=640):
bbs = [BoundingBox(*bb) for bb in bbs]
bbs = BoundingBoxesOnImage(bbs, shape=image.shape)
resize = iaa.Sequential([
iaa.Resize({
"longer-side": image_size,
"shorter-side": "keep-aspect-ratio"
}),
iaa.PadToFixedSize(width=image_size, height=image_size)
])
seq = iaa.Sequential(
[
iaa.Fliplr(0.5), # horizontal flips
# iaa.Crop(percent=(0, 0.05)), # random crops
# Gaussian blur with random sigma between 0 and 0.2.
iaa.GaussianBlur(sigma=(0, 0.2)),
# Strengthen or weaken the contrast in each image.
iaa.LinearContrast((0.8, 1.2)),
# Add gaussian noise.
# For 50% of all images, we sample the noise once per pixel.
# For the other 50% of all images, we sample the noise per pixel AND
# channel. This can change the color (not only brightness) of the
# pixels.
iaa.AdditiveGaussianNoise(
loc=0, scale=(0.0, 0.05 * 255), per_channel=0.5),
# Make some images brighter and some darker.
# In 30% of all cases, we sample the multiplier once per channel,
# which can end up changing the color of the images.
iaa.Multiply((0.8, 1.2), per_channel=0.3),
# Apply affine transformations to each image.
# Scale/zoom them, translate/move them, rotate them and shear them.
iaa.Affine(
scale={
"x": (1, 1.1),
"y": (1, 1.1)
},
# translate_percent={"x": (-0.1, 0.1), "y": (-0.1, 0.1)},
rotate=[90, -90, 180, 88, -92, 178, 182],
shear={
"x": (-5, 5),
"y": (-5, 5)
},
cval=(0, 255)),
],
random_order=True) # apply augmenters in order
image, bbs = resize(image=image, bounding_boxes=bbs)
image, bbs = seq(image=image, bounding_boxes=bbs)
bbs = bbs.remove_out_of_image().clip_out_of_image()
return image, bbs
def aug_sroie_dynamic_1():
"""Perform image augmentation for dynamic input shape"""
aug_list = []
aug_list.append(
augmenters.PadToFixedSize(width=1, height=1536, pad_cval=255))
aug_list.append(augmenters.Affine(translate_px=(-16, 16), cval=255))
aug_list.append(augmenters.Crop(percent=(0.2, 0.3), keep_size=False))
aug_list.append(
augmenters.Resize(size={
"height": 768,
"width": "keep-aspect-ratio"
}))
aug_list.append(augmenters.Fliplr(0.33, name="horizontal_flip"))
aug_list.append(augmenters.Flipud(0.33, name="vertical_flip"))
return aug_list
def train_augmentors(self):
shape_augs = [
iaa.PadToFixedSize(self.data_size[0],
self.data_size[1],
pad_cval=255,
position='center',
deterministic=True),
iaa.Affine(
# scale images to 80-120% of their size, individually per axis
scale={
"x": (0.8, 1.2),
"y": (0.8, 1.2)
},
# translate by -A to +A percent (per axis)
translate_percent={
"x": (-0.01, 0.01),
"y": (-0.01, 0.01)
},
rotate=(-179, 179), # rotate by -179 to +179 degrees
shear=(-5, 5), # shear by -5 to +5 degrees
order=[0], # use nearest neighbour
backend='cv2' # opencv for fast processing
),
iaa.Fliplr(0.5), # horizontally flip 50% of all images
iaa.Flipud(0.5), # vertically flip 20% of all images
iaa.CropToFixedSize(self.input_size[0],
self.input_size[1],
position='center',
deterministic=True),
# iaa.CropToFixedSize(1440, 1440)
]
#
input_augs = [
iaa.OneOf([
iaa.GaussianBlur((0, 3.0)), # gaussian blur with random sigma
iaa.MedianBlur(k=(3, 5)), # median with random kernel sizes
iaa.AdditiveGaussianNoise(loc=0,
scale=(0.0, 0.05 * 255),
per_channel=0.5),
]),
iaa.Sequential([
iaa.Add((-26, 26)),
iaa.AddToHueAndSaturation((-20, 20)),
iaa.LinearContrast((0.75, 1.25), per_channel=1.0),
],
random_order=True),
]
return shape_augs, input_augs
def aug_aocr(args, bg_color=255):
aug_list = []
aug_list.append(augmenters.Affine(
scale={"x": (0.8, 1.0), "y": (0.8, 1.0)}, rotate=(-3, 3), cval=bg_color, fit_output=True),
)
# aug_list.append(augmenters.AllChannelsCLAHE(tile_grid_size_px =(4, 12), per_channel=False))
# aug_list.append(augmenters.AddToHueAndSaturation(value=(-10, 10), per_channel=False))
aug_list.append(augmenters.Resize(
size={"height": args.resize_height, "width": "keep-aspect-ratio"})
)
aug_list.append(augmenters.PadToFixedSize(
height=args.resize_height, width=args.max_img_size, pad_cval=bg_color)
)
aug_list.append(augmenters.CropToFixedSize(
height=args.resize_height, width=args.max_img_size, position="left-top")
)
return aug_list
def infer_augmentors(self):
shape_augs = [
# iaa.Scale(
# {"height": 1280,
# "width" : 1280},
# interpolation='nearest',
# ),
iaa.PadToFixedSize(self.data_size[0],
self.data_size[1],
pad_cval=255,
position='center',
deterministic=True),
iaa.CropToFixedSize(self.input_size[0],
self.input_size[1],
position='center',
deterministic=True),
]
return shape_augs
def __init__(self, data, config, is_train=True):
'Initialization'
self.data = data
self.config = config
self.batch_size = config.BATCH_SIZE
self.n_classes = config.NUM_CLASS
self.h = config.INPUT_SHAPE[0]
self.w = config.INPUT_SHAPE[1]
self.is_train = is_train
self.unique_label = np.unique(data["label"])
self.seq = iaa.Sequential([
iaa.Fliplr(0.5),
iaa.Affine(rotate=(-10, 10)),
iaa.Multiply((0.8, 1.2)),
iaa.PadToFixedSize(width=self.w + 50, height=self.h + 50),
iaa.CropToFixedSize(self.h, self.w, 'normal')
])
self.on_epoch_end()
def aug_sroie_dynamic_2():
"""Bigger and reasonable augmentation"""
aug_list = []
aug_list.append(
augmenters.PadToFixedSize(width=1, height=1536, pad_cval=255))
aug_list.append(augmenters.Affine(translate_px=(-16, 16), cval=255))
# (top, right, bottom, left)
aug_list.append(
augmenters.Crop(percent=((0.25, 0.3), (0.0, 0.1), (0.25, 0.23), (0.0,
0.1)),
keep_size=False))
aug_list.append(
augmenters.Resize(size={
"height": 512,
"width": "keep-aspect-ratio"
}))
aug_list.append(augmenters.Fliplr(0.33, name="horizontal_flip"))
aug_list.append(augmenters.Flipud(0.33, name="vertical_flip"))
return aug_list
def __aug_sequence(self, input_size, target_size):
iw, ih = input_size
i_ratio = iw / ih
tw, th = target_size
t_ratio = tw / th
if i_ratio > t_ratio:
# Input image wider than target, resize width to target width
resize_aug = iaa.Resize({
"width": tw,
"height": "keep-aspect-ratio"
})
else:
# Input image higher than target, resize height to target height
resize_aug = iaa.Resize({
"width": "keep-aspect-ratio",
"height": th
})
pad_aug = iaa.PadToFixedSize(width=tw, height=th, position="center")
seq = iaa.Sequential([resize_aug, pad_aug])
return seq
def __call__(self, img):
np_img = np.asarray(img)
height = np_img.shape[0]
width = np_img.shape[1]
if height >= width:
length = height
else:
length = width
# position is must, because the bbs and images are augmented separately
self.seq = iaa.Sequential([
iaa.PadToFixedSize(width=length,
height=length,
position=self.position)
])
# apply augment
image_aug = self.seq.augment_image(np_img)
return image_aug
def pad_batch_to_fixed_size(batch_images, target_shape, batch_boxes=None):
"""Resize and pad images and boxes to the target shape
Arguments
--------
batch_images: an array of images with shape (H,W,C)
target_shape: a shape of type (H,W,C)
batch_boxes: an array of array with format (xmin, ymin, xmax, ymax)
Returns
------
images_aug: a list of augmented images
boxes_aug: a list of augmented boxes (optional: if boxes is not None)
"""
aug = iaa.Sequential([
iaa.Resize({
"longer-side": target_shape[0],
"shorter-side": "keep-aspect-ratio"
}),
iaa.PadToFixedSize(height=target_shape[0],
width=target_shape[1],
position=(1, 1))
])
images_aug, boxes_aug = __transform_batch(batch_images, aug, batch_boxes)
return images_aug, boxes_aug
def pad_to_fixed_size(image, target_shape, boxes=None):
"""Resize and pad images and boxes to the target shape
Arguments
--------
image: an image with shape (H,W,C)
target_shape: a shape of type (H,W,C)
boxes: an array of format (xmin, ymin, xmax, ymax)
Returns
-------
image_pad: the image padded
boxes_pad: the boxes padded (optional: if boxes is not None)
"""
augmenters = iaa.Sequential([
iaa.Resize({
"longer-side": target_shape[0],
"shorter-side": "keep-aspect-ratio"
}),
iaa.PadToFixedSize(height=target_shape[0],
width=target_shape[1],
position=(1, 1))
])
return __transform(image, augmenters, boxes)
def __init__(self,
images,
labels,
batch_size=16,
image_shape=(256, 512, 1),
do_shuffle_at_epoch_end=True,
length=None,
do_augment=True):
self.number_batches_augmentation = 4
self.labels = labels # array of labels
self.images = images # array of image paths
self.input_shape = image_shape # image dimensions
self.label_shape = (image_shape[0], image_shape[1], 1)
self.length = length
self.batch_size = batch_size # batch size
self.shuffle = do_shuffle_at_epoch_end # shuffle bool
self.augment = do_augment # augment data bool
self.augmenting_pipeline = iaa.Sequential([
iaa.PadToFixedSize(pad_mode="symmetric", width=1024, height=0),
iaa.Affine(shear=(-20, 20)),
iaa.CropToFixedSize(width=512, height=256, position="center"),
iaa.Affine(scale={
"x": (1, 1.5),
"y": (1, 1.5)
},
translate_percent={"x": (-0.4, 0.4)},
mode="symmetric",
cval=[0, 0, 0, 0]),
iaa.PerspectiveTransform(scale=(0, 0.10)),
iaa.CropToFixedSize(width=512, height=256, position="center-top"),
# iaa.PiecewiseAffine(scale=(0.001, 0.01), nb_rows=4, nb_cols=8),
iaa.Lambda(func_images=self.add_background,
func_segmentation_maps=self.convert_segmentations),
])
self.indexes = np.arange(len(self.images))
self.on_epoch_end()