def generate_ridge_data(filename):
"""Generates a ridge data file from a .nii.gz file
Args:
filename (str): Filename to evaluate
Returns:
ndarray: 3-dimensional array of normalized ridge scores
"""
# extract image data from file
background_img = nib.load(filename)
background_data = background_img.get_fdata()
# generate ridge scores
ridge_data = detect_ridges_concurrent(background_data)
# save new image
base_filename = pattern.search(filename).group(1)
ridge_data_filename = f"{base_filename}_ridgeScore.nii.gz"
iu.save_image(ridge_data, ridge_data_filename, background_img.affine)
# normalize and save normalized data
normalized_ridge_data = iu.normalize(ridge_data)
normalized_ridge_data_filename = os.path.join(
f"{base_filename}_ridgeScore_normalized.nii.gz")
iu.save_image(normalized_ridge_data, normalized_ridge_data_filename,
background_img.affine)
# return normalized ridgescores
return ridge_data
async def get_cached_tile_buffer(prefix: str,
var: str,
z: int,
x: int,
y: int,
colorize: str = None):
print("calculate tile metadata")
tile_meta = mercantile.Tile(x=x, y=y, z=z)
print("getting bounds")
min_lon, min_lat, max_lon, max_lat = mercantile.bounds(tile_meta)
print("loading data array")
da = await load_da(prefix, var)
print("indexing tile selection")
data_tile = da.sel(longitude=slice(min_lon, max_lon),
latitude=slice(max_lat, min_lat))
print("getting min/max")
abs_min, abs_max = get_absolute_min_max(prefix, var)
print("Slicing and dicing data")
normalized_tile = resize(
normalize(data_tile, max_val=abs_max, min_val=abs_min))
print("encoding png")
return await encode_as_png(normalized_tile, colorize=colorize)
def step(self, action):
action = int(action)
self._send(CMD_STEP)
self._recv_bytes()
self._send_action(action)
reward, is_done = self._recv_step_json_data()
image = self._recv_state_image()
image = normalize(image)
return image, reward, is_done
def get_clover_dataset(path, size):
'''
クローバー画像を ImageDataSet オブジェクトとして取得する
:param path: positive/negative フォルダを含む保存先パス
:param size: モデルに渡す画像ファイルのサイズ（ピクセル）
:return: ImageDataSet オブジェクト
'''
pos_images = np.array([
np.asarray(img).ravel() for img in normalize(path + '/positive', size)
])
pos_labels = np.full(shape=pos_images.shape[0], fill_value=1)
neg_images = np.array([
np.asarray(img).ravel() for img in normalize(path + '/negative', size)
])
neg_labels = np.full(shape=neg_images.shape[0], fill_value=0)
images = np.concatenate((pos_images, neg_images))
labels = np.concatenate((pos_labels, neg_labels)).reshape(-1, 1)
return ImageDataSet(images, labels)
def main():
for i in range(0, N):
img_name = "../../../input_image/basket/images/%03d.png" % (i)
# img_name = "../test_data/small/%03d_small.png" % (i)
img = cv2.imread(img_name)
gray_image = cv2.cvtColor(img, cv2.cv.CV_BGR2GRAY)
img_list.append(utils.normalize(gray_image))
gradient_map = my_gradient_map.gradient_map(img_list)
cv2.imwrite('output_diffuse.png', utils.denormalize(gradient_map))
cv2.imshow('image', gradient_map)
cv2.waitKey(0)
def generate_batch_variable(image_generator, use_cuda, batch_size):
batch = []
for _ in range(batch_size):
image = next(image_generator)
image = image_utils.normalize(image)
batch.append(image)
batch = np.stack(batch).astype(np.float32)
batch = np.transpose(batch, (0, 3, 1, 2)) # (batch, y, x, channel) -> (batch, channel, y, x)
batch = torch.from_numpy(batch)
if use_cuda:
batch = batch.cuda()
batch = torch.autograd.Variable(batch, requires_grad=False)
return batch
def train_step(image, extractor, optimizer, weights, targets, num_layers,
shift):
with tf.GradientTape() as tape:
# Get current activations
outputs = extractor(image)
loss = style_content_loss(outputs, weights['style'],
weights['content'], weights['frame'],
num_layers['style'], num_layers['content'],
targets['style'], targets['content'],
targets['frame'])
# Add denoise loss
loss += weights['denoise'] * denoise_loss(image, shift)
# Use selected optimizer to descend gradient
grad = tape.gradient(loss, image)
optimizer.apply_gradients([(grad, image)])
# Return the network's new values for the image and clip between 0 and 1
return images.normalize(image)
def generate_fake_image(image, generator_net, use_cuda):
image = image_utils.normalize(image)
image = image[np.newaxis, :, :, :]
image = np.transpose(
image,
(0, 3, 1, 2)) # (batch, y, x, channel) -> (batch, channel, y, x)
image = torch.from_numpy(image)
if use_cuda:
image = image.cuda()
image = torch.autograd.Variable(image, requires_grad=False)
fake = generator_net(image)
fake = fake[0, :, :, :]
if use_cuda:
fake = fake.cpu()
fake = fake.data.numpy()
fake = np.transpose(fake, (1, 2, 0)) # (channel, y, x) -> (y, x, channel)
fake = image_utils.unnormalize(fake[:, :, :])
return fake
def predict(input_path, output_path, model, model_dir, chip_size, channels, grids, batch_size, params):
input_dataset = gdal.Open(input_path)
image = image_utils.load_file(input_path)[:, : , :channels]
image_predicted = np.zeros((image.shape[0], image.shape[1]), dtype=np.int)
params['batch_size'] = batch_size
if model not in models.keys():
raise ValueError('Model ' + model + ' is not supported')
model_fn=models[model]
estimator = tf.estimator.Estimator(model_fn=model_fn, params=params, model_dir=model_dir)
for step in image_utils.get_grids(grids, chip_size):
batch = []
for (x, y, window, original_dimensions) in image_utils.sliding_window(image, step["steps"], step["chip_size"], (chip_size, chip_size)):
if window.shape[0] != chip_size or window.shape[1] != chip_size:
print(window.shape, chip_size)
continue
window_normalized = image_utils.normalize(window)
batch.append({
"window": window_normalized,
"x": x,
"y": y,
"dimensions": original_dimensions
})
if len(batch) >= batch_size:
windows = []
positions = []
dimensions = []
for b in batch:
windows.append(b.get("window"))
positions.append((b.get("x"), b.get("y")))
dimensions.append(b.get("dimensions"))
windows = np.array(windows)
predict_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"data": np.array(windows, dtype=np.float32)},
shuffle=False
)
pred = estimator.predict(input_fn=predict_input_fn)
for window, position, dimension, predict in zip(windows, positions, dimensions, pred):
predict[predict > 0.5] = 1
predict[predict <= 0.5] = 0
predict = image_utils.resize(predict, (dimension[0], dimension[1]), preserve_range=True, anti_aliasing=True).astype(np.int8)
predict = predict.reshape((predict.shape[0], predict.shape[1]))
predicted = image_utils.get_window(image_predicted, position[0], position[1], predict.shape[1], predict.shape[0])
if predict.shape != predicted.shape:
import ipdb; ipdb.set_trace()
try:
image_utils.set_window(image_predicted, np.add(predict, predicted), position[0], position[1])
except Exception as e:
import ipdb; ipdb.set_trace()
batch = []
driver = input_dataset.GetDriver()
output_dataset = driver.Create(output_path, image.shape[1], image.shape[0], 1, gdal.GDT_Int16)
output_dataset.SetGeoTransform(input_dataset.GetGeoTransform())
output_dataset.SetProjection(input_dataset.GetProjection())
output_band = output_dataset.GetRasterBand(1)
output_band.WriteArray(image_predicted.reshape((image_predicted.shape[0], image_predicted.shape[1])), 0, 0)
output_band.FlushCache()
def reset(self):
self._send(CMD_RESET)
image = self._recv_state_image()
image = normalize(image)
return image
image_df = pd.read_csv(images_df_path)
id = max(image_df['id'].tolist())
else:
image_df = pd.DataFrame(columns=['id', 'image'])
id = 0
filename, extension = os.path.basename(image_path).split('.')
filename = filename.replace('_', '-')
for (x, y, window) in sliding_window(image, image_size):
chip = np.array(window[:, :, :image_channels])
chip = np.flipud(chip)
chip = apply_contrast(chip)
chip = normalize(chip)
chip = chip * 255
left = extent[0] + (x * spatial_resolution_x)
right = left + (image_size * spatial_resolution_x)
top = extent[3] - (y * spatial_resolution_y)
bottom = top - (image_size * spatial_resolution_y)
chip_extent = [left, right, bottom, top]
wkt = "POLYGON (({left} {bottom}," \
"{left} {top}," \
"{right} {top}," \
"{right} {bottom}," \
"{left} {bottom}))" \
.format(left=left, right=right, top=top, bottom=bottom)