# compute normalized reconstruction error
file_name_ref = "_".join(["ae_loss", "test_set", flags.object_class]) + ".npy"
file_path_ref = osp.join(conf.ae_dir, "eval", file_name_ref)
if osp.exists(file_path_ref):
ae_loss_per_pc_ref = np.load(file_path_ref)
nre_per_pc = np.divide(ae_loss_per_pc, ae_loss_per_pc_ref)
log_file.write("Normalized reconstruction error: %.9f\n" %
nre_per_pc.mean())
log_file.close()
# use any plotting mechanism, such as matplotlib, to visualize the results
if flags.visualize_results:
i = 0
plot_3d_point_cloud(pc_data_test.point_clouds[i][:, 0],
pc_data_test.point_clouds[i][:, 1],
pc_data_test.point_clouds[i][:, 2],
in_u_sphere=True,
title="Complete input point cloud")
plot_3d_point_cloud(sampled_pc[i][:, 0],
sampled_pc[i][:, 1],
sampled_pc[i][:, 2],
in_u_sphere=True,
title="S-NET sampled points")
plot_3d_point_cloud(reconstructions[i][:, 0],
reconstructions[i][:, 1],
reconstructions[i][:, 2],
in_u_sphere=True,
title="Reconstruction from S-NET points")
log_file = open(osp.join(eval_dir, log_file_name), "w", 1)
log_file.write("Evaluation flags: %s\n" % flags)
log_file.write("Mean ae loss: %.9f\n" % ae_loss_per_pc.mean())
# compute normalized reconstruction error
file_name_ref = "_".join(["ae_loss", "test_set", flags.object_class]) + ".npy"
file_path_ref = osp.join(conf.ae_dir, "eval", file_name_ref)
if osp.exists(file_path_ref):
ae_loss_per_pc_ref = np.load(file_path_ref)
nre_per_pc = np.divide(ae_loss_per_pc, ae_loss_per_pc_ref)
log_file.write("Normalized reconstruction error: %.9f\n" % nre_per_pc.mean())
log_file.close()
# use any plotting mechanism, such as matplotlib, to visualize the results
if flags.visualize_results:
i = 0
plot_3d_point_cloud(
pc_data_test.point_clouds[i],
in_u_sphere=True,
title="Complete input point cloud",
)
plot_3d_point_cloud(
sampled_pc[i][:n_sample_points], in_u_sphere=True, title="SampleNetProgressive sampled points"
)
plot_3d_point_cloud(
reconstructions[i],
in_u_sphere=True,
title="Reconstruction from SampleNetProgressive points",
)
"fps",
"%04d" % n_sample_points,
]) + ".txt")
else:
log_file_name = log_file_name_wo_fps
log_file = open(osp.join(eval_dir, log_file_name), "w", 1)
log_file.write("Mean ae loss: %.9f\n" % loss_per_pc.mean())
# compute normalized reconstruction error
file_path_ref = osp.join(eval_dir, file_name_wo_fps)
if osp.exists(file_path_ref):
loss_per_pc_ref = np.load(file_path_ref)
nre_per_pc = np.divide(loss_per_pc, loss_per_pc_ref)
log_file.write("Normalized reconstruction error: %.3f\n" %
nre_per_pc.mean())
log_file.close()
# use any plotting mechanism, such as matplotlib, to visualize the results
if flags.visualize_results:
i = 0
plot_3d_point_cloud(
pc_data_test.point_clouds[i],
in_u_sphere=True,
title="Complete input point cloud",
)
plot_3d_point_cloud(reconstructions[i],
in_u_sphere=True,
title="Reconstruction from complete input")
n_sample_points, :] # sample form ProgressiveNet points
sorted_pc = ae.sess.run(ae.x_sorted,
feed_dict={ae.x:
complete_pc}) # complete pc sorted by FPS
fps_pc = sorted_pc[:, :n_sample_points, :] # FPS points
# Reconstruct
reconstructed_from_sampled = ae.reconstruct(complete_pc, S=ordered_pc)[0]
reconstructed_from_fps = ae.reconstruct(complete_pc, S=sorted_pc)[0]
# Use any plotting mechanism, such as matplotlib, to visualize the results
i = 0
plot_3d_point_cloud(complete_pc[i][:, 0],
complete_pc[i][:, 1],
complete_pc[i][:, 2],
in_u_sphere=True,
title='Complete input point cloud')
plot_3d_point_cloud(sampled_pc[i][:, 0],
sampled_pc[i][:, 1],
sampled_pc[i][:, 2],
in_u_sphere=True,
title='ProgressiveNet sampled points')
plot_3d_point_cloud(fps_pc[i][:, 0],
fps_pc[i][:, 1],
fps_pc[i][:, 2],
in_u_sphere=True,
title='FPS sampled points')
plot_3d_point_cloud(reconstructed_from_sampled[i][:, 0],
reconstructed_from_sampled[i][:, 1],
reconstructed_from_sampled[i][:, 2],
file_path_ref = osp.join(conf.ae_dir, "eval", file_name_ref)
if osp.exists(file_path_ref):
ae_loss_per_pc_ref = np.load(file_path_ref)
nre_per_pc = np.divide(ae_loss_per_pc, ae_loss_per_pc_ref)
log_file.write("Normalized reconstruction error: %.9f\n" %
nre_per_pc.mean())
log_file.close()
# use any plotting mechanism, such as matplotlib, to visualize the results
if flags.visualize_results:
i = 0
plot_3d_point_cloud(
pc_data_test.point_clouds[i][:, 0],
pc_data_test.point_clouds[i][:, 1],
pc_data_test.point_clouds[i][:, 2],
in_u_sphere=True,
title="Complete input point cloud",
)
plot_3d_point_cloud(sampled_pc[i][:n_sample_points, 0],
sampled_pc[i][:n_sample_points, 1],
sampled_pc[i][:n_sample_points, 2],
in_u_sphere=True,
title="ProgressiveNet sampled points")
plot_3d_point_cloud(
reconstructions[i][:, 0],
reconstructions[i][:, 1],
reconstructions[i][:, 2],
in_u_sphere=True,
title="Reconstruction from ProgressiveNet points",
)