def generate_sss_map(height_map, network_path):
network = load_network(network_path)
rows, cols = height_map.shape
sss_gen_map = np.zeros((rows, cols, 3), dtype=np.uint8)
interval = 2.
height, width = 256, 256
for i in xrange(0, rows - 20, 20):
for j in xrange(0, cols - 20, 20):
height_patch = height_map[i:i + 20, j:j + 20]
height_patch = clip_to_interval(height_patch, interval)
height_patch = cv2.resize(height_patch, (width, height),
interpolation=cv2.INTER_CUBIC)
height_patch = (255. / interval * height_patch).astype(np.uint8)
height_image = cv2.applyColorMap(height_patch, cv2.COLORMAP_JET)
generated = predict_sidescan(network, height_image)
generated = cv2.resize(generated,
dsize=(20, 20),
interpolation=cv2.INTER_LINEAR)
sss_gen_map[i:i + 20, j:j + 20, :] = generated
return sss_gen_map.transpose(1, 0, 2)
def gen_callback(self, bathy_window):
print "Got gen callback!"
generated_window = np.zeros(bathy_window.shape, dtype=np.float32)
interval = 2.
height, width = 256, 256
for i in range(0, 9):
height_patch = bathy_window[:, i*20:(i+1)*20]
height_patch = clip_to_interval(height_patch, interval)
height_patch_debug = height_patch.copy()
height_patch = cv2.resize(height_patch, (width, height), interpolation = cv2.INTER_CUBIC)
height_patch = (255./interval*height_patch).astype(np.uint8)
height_image = cv2.applyColorMap(height_patch, cv2.COLORMAP_JET)
if i < 4:
height_image = np.flip(height_image, axis=1)
generated = predict_sidescan(self.network, height_image)
if i < 4:
generated = np.flip(generated, axis=1)
generated = cv2.resize(generated, dsize=(20, 20), interpolation=cv2.INTER_LINEAR)
generated_window[:, i*20:(i+1)*20] = 1./255.*generated[:, :, 0].astype(np.float32)
#generated_window[:, i*20:(i+1)*20] = 1./interval*height_patch_copy
return generated_window
def normalize_depth_slices(depth_slices):
depth_slices = [
gen_utils.clip_to_interval(d, 2.) / 2. for d in depth_slices
]
for d in depth_slices:
d[d == 1.] = 0.
return depth_slices
def prepare_depth_window(depth_window):
depth_window = gen_utils.clip_to_interval(depth_window, 2.)/2.
depth_window[depth_window==1.] = 0.
b = (depth_window == 0.).astype(np.uint8)
kernel = np.ones((3, 3), np.uint8)
b = cv2.dilate(b, kernel, iterations=1).astype(bool)
depth_window[b] = 0.
#depth_window = cv2.resize(np.rot90(depth_window), (256, 256), interpolation=cv2.INTER_LINEAR)
depth_window = cv2.resize(np.rot90(depth_window), (256, 256), interpolation=cv2.INTER_NEAREST)
depth_window = (255.*np.minimum(np.maximum(depth_window, 0.), 1.)).astype(np.uint8)
depth_window = depth_window[:, :, np.newaxis]
return depth_window
patch_views = patch_draper.sss_patch_views.read_data(sys.argv[1])
#patch_views.append(patch_views[0])
print "Number of patches before sss filtering: ", len(patch_views)
patch_views = filter_sss_coverage(patch_views)
print "Number of patches after sss filtering: ", len(patch_views)
patch_views = filter_map_coverage(patch_views)
print "Number of patches after map filtering: ", len(patch_views)
interval = 2.
for view in patch_views:
view.patch_height = clip_to_interval(view.patch_height, interval)
patch_views = filter_map_nans(patch_views)
print "Number of patches after map nan filtering: ", len(patch_views)
patch_views = filter_map_extremas(patch_views, interval)
print "Number of patches after map extremas filtering: ", len(patch_views)
#for i in range(0, len(patch_views), 10):
# data_vis.plot_patch_views(patch_views[i:i+10])
if not os.path.exists("pix2pix_sonar"):
os.makedirs("pix2pix_sonar")
os.makedirs(os.path.join("pix2pix_sonar", "8"))