def run_benchmark():
dataset_path = os.path.join(pwd, "..", "..", "TestData", "RGBD")
camera_poses = read_trajectory(os.path.join(dataset_path, "odometry.log"))
camera_intrinsics = o3d.camera.PinholeCameraIntrinsic(
o3d.camera.PinholeCameraIntrinsicParameters.PrimeSenseDefault)
volume = o3d.integration.UniformTSDFVolume(
length=4.0,
resolution=512,
sdf_trunc=0.04,
color_type=o3d.integration.TSDFVolumeColorType.RGB8,
)
s = time.time()
for i in range(len(camera_poses)):
color = o3d.io.read_image(
os.path.join(dataset_path, "color", "{0:05d}.jpg".format(i)))
depth = o3d.io.read_image(
os.path.join(dataset_path, "depth", "{0:05d}.png".format(i)))
rgbd = o3d.geometry.RGBDImage.create_from_color_and_depth(
color, depth, depth_trunc=4.0, convert_rgb_to_intensity=False)
volume.integrate(
rgbd,
camera_intrinsics,
np.linalg.inv(camera_poses[i].pose),
)
time_integrate = time.time() - s
s = time.time()
mesh = volume.extract_triangle_mesh()
time_extract_mesh = time.time() - s
s = time.time()
pcd = volume.extract_point_cloud()
time_extract_pcd = time.time() - s
return time_integrate, time_extract_mesh, time_extract_pcd
import cupoch as cph
from trajectory_io import read_trajectory
import numpy as np
if __name__ == "__main__":
camera_poses = read_trajectory("../../testdata/rgbd/odometry.log")
camera_intrinsics = cph.camera.PinholeCameraIntrinsic(
cph.camera.PinholeCameraIntrinsicParameters.PrimeSenseDefault)
volume = cph.integration.UniformTSDFVolume(
length=4.0,
resolution=512,
sdf_trunc=0.04,
color_type=cph.integration.TSDFVolumeColorType.RGB8,
)
for i in range(len(camera_poses)):
print("Integrate {:d}-th image into the volume.".format(i))
color = cph.io.read_image(
"../../testdata/rgbd/color/{:05d}.jpg".format(i))
depth = cph.io.read_image(
"../../testdata/rgbd/depth/{:05d}.png".format(i))
rgbd = cph.geometry.RGBDImage.create_from_color_and_depth(
color, depth, depth_trunc=4.0, convert_rgb_to_intensity=False)
volume.integrate(
rgbd,
camera_intrinsics,
np.linalg.inv(camera_poses[i].pose),
)
print("Extract triangle mesh")
mesh = volume.extract_triangle_mesh()
# Open3D: www.open3d.org
# The MIT License (MIT)
# See license file or visit www.open3d.org for details
# examples/python/pipelines/rgbd_integration_uniform.py
import open3d as o3d
from trajectory_io import read_trajectory
import numpy as np
if __name__ == "__main__":
camera_poses = read_trajectory("../../test_data/RGBD/odometry.log")
camera_intrinsics = o3d.camera.PinholeCameraIntrinsic(
o3d.camera.PinholeCameraIntrinsicParameters.PrimeSenseDefault)
volume = o3d.pipelines.integration.UniformTSDFVolume(
length=4.0,
resolution=512,
sdf_trunc=0.04,
color_type=o3d.pipelines.integration.TSDFVolumeColorType.RGB8,
)
for i in range(len(camera_poses)):
print("Integrate {:d}-th image into the volume.".format(i))
color = o3d.io.read_image(
"../../test_data/RGBD/color/{:05d}.jpg".format(i))
depth = o3d.io.read_image(
"../../test_data/RGBD/depth/{:05d}.png".format(i))
rgbd = o3d.geometry.RGBDImage.create_from_color_and_depth(
color, depth, depth_trunc=4.0, convert_rgb_to_intensity=False)
volume.integrate(
rgbd,
def evaluate(
airsim_traj_path,
meshroom_traj_path,
airsim_ply_path,
meshroom_ply_path,
crop_bbox_path,
alignment_matrix_path,
):
assert os.path.isfile(
airsim_traj_path), f"File not found: '{airsim_traj_path}'"
assert os.path.isfile(
meshroom_traj_path), f"File not found: '{meshroom_traj_path}'"
assert os.path.isfile(
airsim_ply_path), f"File not found: '{airsim_ply_path}'"
assert os.path.isfile(
meshroom_ply_path), f"File not found: '{meshroom_ply_path}'"
assert os.path.isfile(
crop_bbox_path), f"File not found: '{crop_bbox_path}'"
assert os.path.isfile(
alignment_matrix_path), f"File not found: '{alignment_matrix_path}'"
# TODO update TanksAndTemples's python_toolbox/evaluation/run.py and move
# the main "evaluation portion" of the code to the TanksAndTemples class.
os.makedirs(EVALUATION_OUT_FOLDER, exist_ok=False)
# Load reconstruction and according ground-truth
print(meshroom_ply_path)
pcd = o3d.io.read_point_cloud(meshroom_ply_path)
print(airsim_ply_path)
gt_pcd = o3d.io.read_point_cloud(airsim_ply_path)
traj = read_trajectory(meshroom_traj_path) # generated .log file
gt_traj = read_trajectory(airsim_traj_path) # reference .log file
gt_trans = np.loadtxt(
alignment_matrix_path) # alignment matrix (<scene>_trans.txt)
transformation = trajectory_alignment(None, traj, gt_traj, gt_trans)
# Refine alignment by using the actual ground-truth pointcloud
vol = o3d.visualization.read_selection_polygon_volume(crop_bbox_path)
# Registration refinement in 3 iterations
r2 = registration_vol_ds(pcd, gt_pcd, transformation, vol, DTAU, DTAU * 80,
20)
r3 = registration_vol_ds(pcd, gt_pcd, r2.transformation, vol, DTAU / 2,
DTAU * 20, 20)
r = registration_unif(pcd, gt_pcd, r3.transformation, vol, 2 * DTAU, 20)
# Histograms and P/R/F1
precision, recall, fscore, *histograms_data = EvaluateHisto(
SCENE,
filename_mvs=EVALUATION_OUT_FOLDER,
source=pcd,
target=gt_pcd,
trans=r.transformation,
crop_volume=vol,
voxel_size=DTAU / 2,
threshold=DTAU,
plot_stretch=5,
)
# XXX ^^^^ move to a specific function
print("==============================")
print("evaluation result : %s" % SCENE)
print("==============================")
print("distance tau : %.3f" % DTAU)
print("precision : %.4f" % precision)
print("recall : %.4f" % recall)
print("f-score : %.4f" % fscore)
print("==============================")
# Plotting
edges_source, cum_source, edges_target, cum_target = histograms_data
plot_graph(
SCENE,
mvs_outpath=EVALUATION_OUT_FOLDER,
fscore=fscore,
edges_source=edges_source,
cum_source=cum_source,
edges_target=edges_target,
cum_target=cum_target,
plot_stretch=5,
dist_threshold=DTAU,
)