def cache_train_16():
print('num_train_images =', train_wkt['ImageId'].nunique())
train_shapes = shapes[shapes['image_id'].isin(
train_wkt['ImageId'].unique())]
np.save('../data/train_ids.npy', train_shapes['image_id'])
min_train_height = train_shapes['height'].min()
min_train_width = train_shapes['width'].min()
ids = []
i = 0
for image_id in tqdm(sorted(train_wkt['ImageId'].unique())):
image = extra_functions.read_image_16(image_id)
_, height, width = image.shape
img = image[:, :min_train_height, :min_train_width]
img_mask = extra_functions.generate_mask(
image_id,
height,
width,
num_mask_channels=num_mask_channels,
train=train_wkt)[:, :min_train_height, :min_train_width]
np.save('../data/data_files/{}_img.npy'.format(image_id), img)
np.save('../data/data_files/{}_mask.npy'.format(image_id), img_mask)
ids += [image_id]
i += 1
def cache_train_16():
print('num_train_images =', train_wkt['ImageId'].nunique())
train_shapes = shapes[shapes['image_id'].isin(
train_wkt['ImageId'].unique())]
min_train_height = train_shapes['height'].min()
min_train_width = train_shapes['width'].min()
num_train = train_shapes.shape[0]
image_rows = min_train_height
image_cols = min_train_width
num_channels = 16
num_mask_channels = 10
f = h5py.File(os.path.join(data_path, 'train_16.h5'),
'w',
compression='blosc:lz4',
compression_opts=9)
imgs = f.create_dataset('train',
(num_train, num_channels, image_rows, image_cols),
dtype=np.float16)
imgs_mask = f.create_dataset(
'train_mask', (num_train, num_mask_channels, image_rows, image_cols),
dtype=np.uint8)
ids = []
i = 0
for image_id in tqdm(sorted(train_wkt['ImageId'].unique())):
image = extra_functions.read_image_16(image_id)
_, height, width = image.shape
imgs[i] = image[:, :min_train_height, :min_train_width]
imgs_mask[i] = extra_functions.generate_mask(
image_id,
height,
width,
num_mask_channels=num_mask_channels,
train=train_wkt)[:, :min_train_height, :min_train_width]
ids += [image_id]
i += 1
# fix from there: https://github.com/h5py/h5py/issues/441
f['train_ids'] = np.array(ids).astype('|S9')
f.close()
def cache_train_16():
num_channels = 13
num_mask_channels = 1
for label_csv in tqdm(sorted(os.listdir(label_path))):
# collect: type int, choose the COLLECT to transform
if not COLLECTS[collect] in label_csv:
continue
train_wkt = pd.read_csv(
os.path.join(image_path, 'summaryData_Train_2', label_csv))
print("image number in collect : ", label_csv, ' ',
train_wkt['ImageId'].nunique(), label_csv.rstrip('.csv'))
num_train = train_wkt['ImageId'].nunique()
f = h5py.File(
os.path.join(output_train_path,
label_csv.rstrip('.csv') + '.h5'), 'w')
imgs = f.create_dataset('train', (num_train, num_channels, 900, 900),
dtype=np.float16)
imgs_mask = f.create_dataset('train_mask',
(num_train, num_mask_channels, 900, 900),
dtype=np.uint8)
i = 0
print("<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>")
print("normalization_value: ", normalization_value[collect])
print("<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>")
for image_id in tqdm(sorted(train_wkt['ImageId'].unique())):
# if image_id in imperfect_images:
# print("imperfect pictures: ", image_id)
# continue
imgs[i] = extra_functions.read_image_16(
image_path, label_csv.rstrip('_Train.csv'), image_id,
normalization_value[collect])
imgs_mask[i] = extra_functions.generate_mask_for_image(
image_id, train_wkt)
# mask = imgs_mask[i, 0, :, :]
# print(mask.shape)
# ax1 = plt.subplot(111)
# ax1.set_title(image_id)
# ax1.imshow(mask)
# plt.show()
i = i + 1
f.close()
def cache_train_16():
print('num_train_images =', train_wkt['ImageId'].nunique())
train_shapes = shapes[shapes['image_id'].isin(
train_wkt['ImageId'].unique())]
np.save('../data/train_ids.npy', train_shapes['image_id'])
min_train_height = train_shapes['height'].min()
min_train_width = train_shapes['width'].min()
# num_train = train_shapes.shape[0]
# image_rows = min_train_height
# image_cols = min_train_width
# num_channels = 16
#f = h5py.File(os.path.join(data_path, 'train_16.h5'), 'w')
#imgs = f.create_dataset('train', (num_train, num_channels, image_rows, image_cols), dtype=np.float16)
#imgs_mask = f.create_dataset('train_mask', (num_train,image_rows, image_cols), dtype=np.uint8)
ids = []
i = 0
for image_id in tqdm(sorted(train_wkt['ImageId'].unique())):
image = extra_functions.read_image_16(image_id)
_, height, width = image.shape
img = image[:, :min_train_height, :min_train_width]
img_mask = extra_functions.generate_mask(
image_id,
height,
width,
num_mask_channels=num_mask_channels,
train=train_wkt)[:, :min_train_height, :min_train_width]
np.save('../data/data_files/{}_img.npy'.format(image_id), img)
np.save('../data/data_files/{}_mask.npy'.format(image_id), img_mask)
ids += [image_id]
i += 1
#%%
def mask2poly(predicted_mask, threashold, x_scaler, y_scaler):
polygons = extra_functions.mask2polygons_layer(
predicted_mask[0] > threashold, epsilon=0, min_area=5)
polygons = shapely.affinity.scale(polygons,
xfact=1.0 / x_scaler,
yfact=1.0 / y_scaler,
origin=(0, 0, 0))
return shapely.wkt.dumps(polygons.buffer(2.6e-5))
#%%
for image_id in tqdm(test_ids):
image = extra_functions.read_image_16(image_id)
file_name = '{}.tif'.format(image_id)
image_3 = tiff.imread(os.path.join(three_band_path, file_name))
image_3 = np.transpose(image_3, (1, 2, 0))
image_3 = image_3 / 2047 * 255
image_3 = np.array(image_3, dtype=np.uint8)
H = image.shape[1]
W = image.shape[2]
x_max, y_min = extra_functions._get_xmax_ymin(image_id)
predicted_mask = extra_functions.make_prediction_cropped(
model,
image,
initial_size=(128, 128),
print("<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>")
print("normalization_value: ", normalization_value[index])
print("<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>")
for i in range(n_ims):
image_id = im_list[i]
print(
"<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>"
)
print("predict image id: ", image_id)
print(
"<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>"
)
image = extra_functions.read_image_16(test_path, collect_name,
image_id,
normalization_value[index])
predicted_mask = extra_functions.make_prediction_cropped(
model,
image,
initial_size=(patch_width, patch_height),
final_size=(patch_width - 32, patch_height - 32),
num_masks=1,
num_channels=channels)
image_v = flip_axis(image, 1)
predicted_mask_v = extra_functions.make_prediction_cropped(
model,
image_v,
initial_size=(patch_width, patch_height),
