def __init__(self, sample_df, data_path, output_size=512, mask_img=True):
"""
Constructor of the dataset.
----------
INPUT
|---- sample_df (pandas.DataFrame) the dataframe containing the
| samples' filenames, labels, (semi-labels and mask-filename).
|---- data_path (str) the path to the data filenames specified in the sample_df
|---- output_size (int) the size of the output squared image.
|---- mask_img (bool) whether to mask image.
"""
data.Dataset.__init__(self)
self.sample_df = sample_df
self.data_path = data_path
self.transform = tf.Compose(
tf.Grayscale(), tf.AutoContrast(cutoff=1),
tf.RandomHorizontalFlip(p=0.5),
tf.RandomScaling(scale_range=(0.8, 1.2)),
tf.RandomRotation(degree_range=(-45, 45)),
tf.ResizeMax(max_len=output_size), tf.PadToSquare(),
tf.RandomCropResize((output_size, output_size),
scale=(0.4, 1.0),
ratio=(4. / 5., 5. / 4.)),
tf.ColorDistorsion(s=0.5),
tf.GaussianBlur(p=0.5, sigma=(0.1, 2.0)), tf.MinMaxNormalization(),
tf.MaskImage(mask_img), tf.ToTorchTensor())
def __init__(self,
sample_df,
data_path,
mask_img=True,
output_size=512,
data_augmentation=True):
"""
Constructor of the dataset.
----------
INPUT
|---- sample_df (pandas.DataFrame) the dataframe containing the
| samples' filenames, labels, (semi-labels and mask-filename).
|---- data_path (str) the path to the data filenames specified in the sample_df
|---- mask_img (bool) whether to mask the image.
|---- output_size (int) the size of the output squared image.
|---- data_augmentation (bool) whether to perform data augmentation.
"""
data.Dataset.__init__(self)
self.sample_df = sample_df
self.data_path = data_path
if data_augmentation:
self.transform = tf.Compose(tf.Grayscale(), \
tf.AutoContrast(cutoff=1), \
tf.RandomHorizontalFlip(p=0.5), \
tf.RandomVerticalFlip(p=0.5), \
tf.RandomBrightness(lower=0.8, upper=1.2), \
tf.RandomScaling(scale_range=(0.8,1.2)), \
tf.RandomRotation(degree_range=(-20,20)), \
tf.ResizeMax(output_size), \
tf.PadToSquare(), \
tf.MinMaxNormalization(), \
tf.MaskImage(mask_img), \
tf.ToTorchTensor())
else:
self.transform = tf.Compose(tf.Grayscale(), \
tf.AutoContrast(cutoff=1), \
tf.ResizeMax(output_size), \
tf.PadToSquare(), \
tf.MinMaxNormalization(), \
tf.MaskImage(mask_img), \
tf.ToTorchTensor())
range=(df[f'AD_scores_{i}'].min(), df[f'AD_scores_{i}'].max()),
color='limegreen',
alpha=0.4)
ax.set_title(part.title(), fontsize=12)
ax.ticklabel_format(style='sci', axis='x', scilimits=(0, 0))
fig.savefig(FIGURE_PATH + 'score_distribution.pdf',
dpi=dpi,
bbox_inches='tight')
plt.show()
#%%#############################################################################
# 10 highest scores by body part #
################################################################################
img_transform = tf.Compose(tf.Grayscale(), tf.ResizeMax(max_len=512),
tf.PadToSquare(), tf.ToTorchTensor())
j = 1
for label, low_ad in zip([1, 1, 0, 0], [False, True, False, True]):
fig, axs = plt.subplots(7,
10,
figsize=(20, 14),
gridspec_kw={
'hspace': 0.25,
'wspace': 0.0
})
if transparent: fig.set_alpha(0.0)
for ax_row, part in zip(axs, BodyPart_list):
df_high = df[(df.body_part == part)
& (df.abnormal_XR == label)].sort_values(