def convert_and_save(args, i):
ptc = load_velo_scan(osp.join(args.input_path, "{:06d}.bin".format(i)))
calib = Calibration(osp.join(args.calib_path, "{:06d}.txt".format(i)))
img = load_image(osp.join(args.image_path, "{:06d}.png".format(i)))
ptc = get_ptc_in_image(ptc, calib, img)
depth_map = get_depth_map(ptc, calib, img)
np.save(osp.join(args.output_path, "{:06d}".format(i)), depth_map)
def gen_sparse_points(data_idx, args):
calib = Calibration(osp.join(args.calib_path, "{:06d}.txt".format(data_idx)))
pc_velo = load_velo_scan(osp.join(args.ptc_path, "{:06d}.bin".format(data_idx)))
img = load_image(osp.join(args.image_path, "{:06d}.png".format(data_idx)))
img_height, img_width, img_channel = img.shape
_, _, valid_inds_fov = get_lidar_in_image_fov(
pc_velo[:, :3], calib, 0, 0, img_width, img_height, True)
pc_velo = pc_velo[valid_inds_fov]
if(args.div > 1):
valid_inds = (pc_velo[:, 1] < 50) & \
(pc_velo[:, 1] >= -50) & \
(pc_velo[:, 2] < 1.5) & \
(pc_velo[:, 2] >= -2.5)
else:
valid_inds = (pc_velo[:, 0] < 120) & \
(pc_velo[:, 0] >= 0) & \
(pc_velo[:, 1] < 50) & \
(pc_velo[:, 1] >= -50) & \
(pc_velo[:, 2] < 1.5) & \
(pc_velo[:, 2] >= -2.5)
pc_velo = pc_velo[valid_inds]
if args.fill_in_map_dir is not None and (args.fill_in_spec is not None or args.fill_in_slice is not None):
fill_in_line = np.load(os.path.join(args.fill_in_map_dir, "{:06d}.npy".format(data_idx)))
else:
fill_in_line = None
if args.store_line_map_dir is not None:
depth_map_lines, ptc = pto_ang_map(pc_velo, H=args.H, W=args.W, slice=args.slice, div=args.div,\
line_spec=args.line_spec, argo=args.argo, get_lines=True,\
fill_in_line=fill_in_line, fill_in_spec=args.fill_in_spec,
fill_in_slice=args.fill_in_slice)
np.save(osp.join(args.store_line_map_dir,
"{:06d}".format(data_idx)), depth_map_lines)
return ptc
else:
return pto_ang_map(pc_velo, H=args.H, W=args.W, slice=args.slice, div=args.div,\
line_spec=args.line_spec, argo=args.argo, get_lines=False,
fill_in_line=fill_in_line, fill_in_spec=args.fill_in_spec,
fill_in_slice=args.fill_in_slice)
def gen_sparse_points(data_idx, args):
calib = Calibration(
osp.join(args.calib_path, "{:06d}.txt".format(data_idx)))
pc_velo = load_velo_scan(
osp.join(args.ptc_path, "{:06d}.bin".format(data_idx)))
# print('pc_velo', pc_velo.shape)
img = load_image(osp.join(args.image_path, "{:06d}.png".format(data_idx)))
img_height, img_width, img_channel = img.shape
# print('img', img.shape)
_, _, valid_inds_fov = get_lidar_in_image_fov(pc_velo[:, :3], calib, 0, 0,
img_width, img_height, True)
pc_velo = pc_velo[valid_inds_fov]
valid_inds = (pc_velo[:, 0] < 120) & \
(pc_velo[:, 0] >= 0) & \
(pc_velo[:, 1] < 50) & \
(pc_velo[:, 1] >= -50) & \
(pc_velo[:, 2] < 1.5) & \
(pc_velo[:, 2] >= -2.5)
pc_velo = pc_velo[valid_inds]
# print('pc_velo', pc_velo.shape)
if args.fill_in_map_dir is not None and (args.fill_in_spec is not None or
args.fill_in_slice is not None):
fill_in_line = np.load(
os.path.join(args.fill_in_map_dir, "{:06d}.npy".format(data_idx)))
else:
fill_in_line = None
if args.store_line_map_dir is not None:
depth_map_lines, ptc = pto_ang_map(data_idx,
pc_velo,
H=args.H,
W=args.W,
slice=args.slice,
line_spec=args.line_spec,
get_lines=True,
fill_in_line=fill_in_line,
fill_in_spec=args.fill_in_spec,
fill_in_slice=args.fill_in_slice)
np.save(osp.join(args.store_line_map_dir, "{:06d}".format(data_idx)),
depth_map_lines)
return ptc
else:
depth_map = pto_ang_map(data_idx,
pc_velo,
H=args.H,
W=args.W,
slice=args.slice,
line_spec=args.line_spec,
get_lines=False,
fill_in_line=fill_in_line,
fill_in_spec=args.fill_in_spec,
fill_in_slice=args.fill_in_slice)
# should save a 3D array in size (H, M, 4)
np.save(osp.join(args.output_path, "{:06d}".format(data_idx)),
depth_map)
return depth_map
ptc = load_velo_scan(osp.join(args.input_path, "{:06d}.bin".format(i)))
calib = Calibration(osp.join(args.calib_path, "{:06d}.txt".format(i)))
img = load_image(osp.join(args.image_path, "{:06d}.png".format(i)))
ptc = get_ptc_in_image(ptc, calib, img)
depth_map = get_depth_map(ptc, calib, img)
np.save(osp.join(args.output_path, "{:06d}".format(i)), depth_map)
if __name__ == "__main__":
if not os.path.exists(args.output_path):
os.makedirs(args.output_path)
if args.i is not None:
i = args.i
ptc = load_velo_scan(osp.join(args.input_path, "{:06d}.bin".format(i)))
calib = Calibration(osp.join(args.calib_path, "{:06d}.txt".format(i)))
img = load_image(osp.join(args.image_path, "{:06d}.png".format(i)))
ptc = get_ptc_in_image(ptc, calib, img)
depth_map = get_depth_map(ptc, calib, img)
np.save(osp.join(args.output_path, "{:06d}".format(i)), depth_map)
else:
with open(args.split_file) as f:
idx_list = [
int(x.strip()) for x in f.readlines() if len(x.strip()) > 0
]
pbar = tqdm(total=len(idx_list))
def update(*a):
pbar.update()
pool = Pool(args.threads)
res = []
def get_right_image(self, idx):
img_filename = os.path.join(self.right_image_dir, '%06d.png' % (idx))
return utils.load_image(img_filename)