def __getitem__(self, idx):
if self.mode=='train':
item = self.data[idx]
img_path = self.path + '%s.png'%item['img_id']
img = plt.imread(img_path)
width, height = img.shape
img = cv2.resize(img, (self.IMG_SIZE, self.IMG_SIZE))
img = np.expand_dims(img, 0)
cnt_masks = item['cnt_masks']
masks_in_rle = item['masks']
if cnt_masks==1:
mask = rle2mask(masks_in_rle[0], width, height).T
elif cnt_masks>1: #v1: just simply merge overlapping masks to get union of masks
masks = []
for mask_in_rle in masks_in_rle:
mask = rle2mask(mask_in_rle, width, height).T
masks.append(mask)
mask = (np.array(masks).sum(axis=0)>=1).astype(np.int)#mask as 1 if at least one of the mask is 1
else:
mask = np.zeros((self.IMG_SIZE, self.IMG_SIZE))
mask = cv2.resize(mask.astype(np.float), (self.IMG_SIZE, self.IMG_SIZE))
mask = np.expand_dims(mask, 0)
##augmentation
if self.augmentation:
img, mask = do_augmentation(img, mask)
return img, mask
elif self.mode=='test':
img_id = self.img_id_list[idx]
img_path = self.path + '%s.png'%img_id
img = plt.imread(img_path)
width, height = img.shape
img = cv2.resize(img, (self.IMG_SIZE, self.IMG_SIZE))
img = np.expand_dims(img, 0)
return img
def csv2mask(in_path, out_dir, resize=None, v=0):
"""
Converts given .csv annotations, creates masks in out_dir
:param in_path: str: file path to `.csv`
:param out_dir: str: path to out dir where to save .png images
:param resize: tuple(int, int): if provided saves
resized images
:param v: int: verbosity, if 1 shows `tqdm`
:return: NoneType: None object
"""
df = pd.read_csv(in_path, dtype=str, sep=", ")
df = df.groupby('ImageId')["EncodedPixels"].apply(list)
it = df.items()
if v:
it = tqdm(it)
for image_id, row in it:
mask = blank_mask(WIDTH, HEIGHT)
for i, rle_mask in enumerate(row, 1):
if rle_mask == "-1":
continue
i_mask = rle2mask(rle_mask, WIDTH, HEIGHT)
mask[i_mask > 125] = i
out_fname = "{}.png".format(image_id)
out_fp = os.path.join(out_dir, out_fname)
if resize:
mask = cv2.resize(mask, resize)
cv2.imwrite(out_fp, mask)
def test():
df = pd.read_csv('../input/train-rle.csv', header=None, index_col=0)
Y_train = np.zeros((1024, 1024, 1))
for x in df.loc['1.2.276.0.7230010.3.1.4.8323329.307.1517875162.311533',
1]:
# plt.imshow(rle2mask(x, 1024, 1024).T)
# plt.show()
Y_train = Y_train + np.expand_dims(rle2mask(x, 1024, 1024).T, axis=2)
print(np.shape(Y_train))
Y_train = np.squeeze(Y_train, 2)
Y_train = np.where(Y_train > 0, 255, 0)
plt.imshow(Y_train)
plt.show()
accu_bug_best = pd.read_csv(
"./submissions/submission_private_accum_bug_tta_best.csv", dtype=str)
best_new_dfs = [accu_bug_best, best_align, best_align_div]
dfs = [our_best] + best_new_dfs
out_values = []
for img_id in tqdm(our_best.ImageId.unique()):
or_between = None
skip = False
for i, df in enumerate(dfs):
val = df[df.ImageId == img_id]["EncodedPixels"].values[0]
if val.strip() == "-1":
skip = True
break
mask = (rle2mask(val, 1024, 1024) > 0).astype(np.uint8)
if or_between is None:
or_between = np.zeros_like(mask, dtype=np.bool)
or_between |= mask > 0.5
if skip:
out_values.append([img_id, "-1"])
else:
out_values.append([img_id, better_mask2rle(or_between * 1 * 255)])
out_best_values = out_values.copy()
out_values_df = pd.DataFrame(out_best_values,
columns=['ImageId', 'EncodedPixels'])
print((out_values_df.EncodedPixels == "-1").sum())
out_values_df.to_csv("private_merge_best.csv", index=None)
def get_mask(encode, height, width):
mask = rle2mask(encode[0], height, width)
for e in encode[1:]:
mask += rle2mask(e, height, width)
print(mask)
return mask.T
def dcm2jpg(dcm_path,
save_jpg_path,
csv_path=None,
save_mask_path=None,
show_info_pic=False):
count_no_mask = 0
if not os.path.exists(save_jpg_path):
os.makedirs(save_jpg_path)
# if has .csv
if csv_path:
# header=None : Instead of using the first row as a column index, the index is automatically generated
# index_col=0 : Use the first column as an index column
df = pd.read_csv(csv_path, header=None, index_col=0)
if not os.path.exists(save_mask_path):
os.makedirs(save_mask_path)
for file_path in glob.glob(dcm_path):
# convert dcm to jpg
dataset = pydicom.dcmread(file_path)
if show_info_pic:
show_dcm_info(dataset, file_path)
plot_pixel_array(dataset)
print(file_path)
if not csv_path:
misc.imsave(
os.path.join(save_jpg_path,
file_path.split('/')[-1][:-4] + '.jpg'),
dataset.pixel_array)
# if has .csv, convert csv to mask matrix
# notice, .loc[row_index, col_index], and if the mask matrix must be transposed
if csv_path:
try:
# there is no mask
if '-1' in df.loc[file_path.split('/')[-1][:-4],
1] or ' -1' in df.loc[
file_path.split('/')[-1][:-4], 1]:
print('no mask')
mask = np.zeros((1024, 1024))
else:
# if has one mask
if type(df.loc[file_path.split('/')[-1][:-4], 1]) == str:
print('one mask')
mask = rle2mask(
df.loc[file_path.split('/')[-1][:-4], 1], 1024,
1024).T
mask = np.where(mask > 0, 255, 0)
# more than one mask
else:
print('more than one mask')
mask = np.zeros((1024, 1024))
for x in df.loc[file_path.split('/')[-1][:-4], 1]:
mask = mask + rle2mask(x, 1024, 1024).T
mask = np.where(mask > 0, 255, 0)
print('save')
misc.imsave(
os.path.join(save_jpg_path,
file_path.split('/')[-1][:-4] + '.jpg'),
dataset.pixel_array)
misc.imsave(
os.path.join(save_mask_path,
file_path.split('/')[-1][:-4] + '.png'), mask)
except KeyError:
print("Key" + file_path.split('/')[-1][:-4] + ",without mask")
count_no_mask += 1
print('count_no_mask:', count_no_mask)