def save_result(train_image, train_polygon, train_y_coords, filepath):
# Image with polygons overlaid
im_res = train_image.shape[0]
image = (train_image - INPUT_DYNAMIC_RANGE[0]) / (INPUT_DYNAMIC_RANGE[1] -
INPUT_DYNAMIC_RANGE[0])
train_polygon = train_polygon * im_res
train_y_coords = train_y_coords * im_res
plt.cla()
fig = plt.imshow(image)
polygon_utils.plot_polygon(train_polygon,
color="#28ff0288",
draw_labels=False)
polygon_utils.plot_polygon(train_y_coords,
color="#ff8800",
draw_labels=False)
plt.margins(0)
plt.axis('off')
fig.axes.get_xaxis().set_visible(False)
fig.axes.get_yaxis().set_visible(False)
plt.savefig(filepath + ".png", bbox_inches='tight', pad_inches=0)
# Save polygons
train_polygon_array = np.array(train_polygon, dtype=np.float16)
train_y_coords_array = np.array(train_y_coords, dtype=np.float16)
np.save(filepath + ".gt.npy", train_polygon_array)
np.save(filepath + ".pred.npy", train_y_coords_array)
def plot_polygons(polygons, color):
# print("plot_polygons(polygons, color)") # TODO: remove
for i, polygon in enumerate(polygons):
# Remove coordinates after nans
indexes_of_nans = np.where(np.isnan(polygon[:, 0]))[0]
if len(indexes_of_nans):
polygon_nans_crop = polygon[:indexes_of_nans[-1], :]
polygon_utils.plot_polygon(polygon_nans_crop, color=color, draw_labels=False, indexing="ij")
else:
polygon_utils.plot_polygon(polygon, color=color, draw_labels=False, indexing="ij")
def plot_results(figure_index, image_batch, polygon_batch, y_image_batch):
im_res = image_batch[0].shape[0]
image = (image_batch[0] - INPUT_DYNAMIC_RANGE[0]) / (
INPUT_DYNAMIC_RANGE[1] - INPUT_DYNAMIC_RANGE[0])
y_image = y_image_batch[0]
train_polygon = polygon_batch[0] * im_res
plt.figure(figure_index)
plt.cla()
plt.imshow(image)
plt.imshow(y_image[:, :, 0], alpha=0.5, cmap="gray")
polygon_utils.plot_polygon(train_polygon, label_direction=1)
plt.draw()
plt.pause(0.001)
def plot_results(figure_index, train_image_batch, train_polygon_batch,
train_y_coords_batch):
im_res = train_image_batch[0].shape[0]
image = (train_image_batch[0] - INPUT_DYNAMIC_RANGE[0]) / (
INPUT_DYNAMIC_RANGE[1] - INPUT_DYNAMIC_RANGE[0])
train_polygon = train_polygon_batch[0] * im_res
train_y_coords = train_y_coords_batch[0] * im_res
plt.figure(figure_index)
plt.cla()
plt.imshow(image)
polygon_utils.plot_polygon(train_polygon, label_direction=1)
polygon_utils.plot_polygon(train_y_coords, label_direction=-1)
plt.draw()
plt.pause(0.001)
# The op for initializing the variables.
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
with tf.Session() as sess:
sess.run(init_op)
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
# Let's read off 3 batches just for example
for i in range(1):
imgs, polys, raster_polys = sess.run(
[image, polygon, raster_polygon])
print(imgs.min(), imgs.max())
print(polys.min(), polys.max())
print(raster_polys.min(), raster_polys.max())
for img, poly, raster_poly in zip(imgs, polys, raster_polys):
img = (img - dynamic_range[0]) / (dynamic_range[1] -
dynamic_range[0])
poly = poly * im_res
io.imshow(img)
io.imshow(raster_poly[:, :, 0], alpha=0.5, cmap="gray")
polygon_utils.plot_polygon(poly, label_direction=1)
io.show()
coord.request_stop()
coord.join(threads)