def __init__(self, file_name, mask_files, file_type, patch_type, auto_save_patch = True):
self._cfg = config_fun.config()
self._file_name = file_name
self._mask_files = mask_files
self._auto_save_patch = auto_save_patch
self._file_type = file_type #train or val
self._patch_type = patch_type #postive or negative
self._neg_start_idx = 3 #neglect!!!
self._pos_start_idx = self._neg_start_idx + self._cfg.num_neg_classes #neglect!!! 3+1=4
self._img = slide_fun.AllSlide(self._file_name)
self._merged_mask_size = np.ceil(np.array(self._img.level_dimensions[0])/self._cfg.rescaled_times).astype(np.int)
self._merged_mask = None
self._merged_mask_level = None
self._final_level = self._get_final_level(self._cfg.target_level)
self._rescaled_times = self._get_rescaled_times_of_final_level(self._final_level)
print('level %d rescaled times campared with level0: %d'%(self._final_level, self._rescaled_times))
self._times_target_level_divide_rescaled_mask = int(self._cfg.rescaled_times/self._rescaled_times)
self._min_patch_size = int(self._cfg.patch_size/self._times_target_level_divide_rescaled_mask)
def __init__(self, data_type='train', frac=1, file_name=None):
cfg = config_fun.config()
self._data_type = data_type
self._compose = patch_preprocess_fun.get_train_val_compose()
if file_name is None:
self._data, self._label, self._name = hdf5_fun.get_all_data_label_name(
cfg, data_type=data_type, frac=frac)
else:
self._data, self._label, self._name = hdf5_fun.h5_extract_data_label_name(
cfg.patch_size, file_name)
assert self._data.shape[0] == self._label.shape[0]
def get_gm_compose(input_size=None):
cfg = config_fun.config()
if input_size is None: input_size = get_input_size(cfg)
info = get_mean_std(cfg)
normalize = transforms.Normalize(mean=info['mean'], std=info['std'])
compose = transforms.Compose([
transforms.Resize(input_size),
transforms.ToTensor(),
normalize,
])
return compose
def get_h5_compose():
cfg = config_fun.config()
input_size = get_input_size(cfg)
info = get_mean_std(cfg)
normalize = transforms.Normalize(mean=info['mean'], std=info['std'])
compose = transforms.Compose([
transforms.ToPILImage(),
transforms.RandomResizedCrop(input_size),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
normalize,
])
return compose
def __init__(self, file_name, mask_files, file_type, patch_type, auto_save_patch = True):
self._cfg = config_fun.config()
self._file_name = file_name
self._mask_files = mask_files
self._auto_save_patch = auto_save_patch
self._file_type = file_type
self._patch_type = patch_type
self._neg_start_idx = 3
self._pos_start_idx = self._neg_start_idx + self._cfg.num_neg_classes
self._img = slide_fun.AllSlide(self._file_name)
self._max_mask = None
self._max_mask_size = np.ceil(np.array(self._img.level_dimensions[0])/self._cfg.max_frac).astype(np.int)
self._max_mask_level = None
self._min_patch_size = int(self._cfg.patch_size/self._cfg.min_frac)
def convert_patch_to_hdf5():
cfg = config_fun.config()
train_pos_patches = glob.glob(
os.path.join(cfg.patch_save_folder, 'train', 'pos', '*' + cfg.img_ext))
train_neg_patches = glob.glob(
os.path.join(cfg.patch_save_folder, 'train', 'neg', '*' + cfg.img_ext))
val_pos_patches = glob.glob(
os.path.join(cfg.patch_save_folder, 'val', 'pos', '*' + cfg.img_ext))
val_neg_patches = glob.glob(
os.path.join(cfg.patch_save_folder, 'val', 'neg', '*' + cfg.img_ext))
train_patches = _add_label(train_pos_patches, train_neg_patches)
val_patches = _add_label(val_pos_patches, val_neg_patches)
print('processing train patches~')
_precoss_patches(cfg, train_patches, 'train')
print('processing validation patches~')
_precoss_patches(cfg, val_patches, 'val')
def generate_patch(auto_save_patch=True):
cfg = config_fun.config()
with open(cfg.split_file) as f:
split_data = json.load(f)
train_data = filter(
lambda item: item['info'] == 'train_tumor' or item['info'] ==
'train_normal', split_data)
val_data = filter(
lambda item: item['info'] == 'val_tumor' or item['info'] ==
'val_normal', split_data)
# test_data = filter(lambda item: item['info'] == 'test_tumor' or
# item['info'] == 'test_normal', split_data)
_prepare_data(cfg, train_data, 'train', auto_save_patch=auto_save_patch)
_prepare_data(cfg, val_data, 'val', auto_save_patch=auto_save_patch)
train_patch = get_coors(cfg, 'train')
val_patch = get_coors(cfg, 'val')
print('train file %d' % len(train_patch))
print('val file %d' % len(val_patch))
def random_vis_hdf5():
cfg = config_fun.config()
print('vis train hdf5 ~')
_random_vis_hdf5(cfg, 'train')
print('vis validation hdf5 ~')
_random_vis_hdf5(cfg, 'val')