Add Point Cloud to Mesh Comparison Tool with Heatmap Visualization

CHANGELOG.md

@@ -71,6 +71,7 @@
 -   Fix linker error "library limit of 65535 objects exceeded" with Ninja generator on MSVC (PR #7335)
 -   Download tarballs instead of Git repos for "3rdparty/uvatlas" (PR #7371)
 -   macOS x86_64 not longer supported, only macOS arm64 is supported.
+-   Add point cloud to mesh/cloud comparison example with RANSAC global registration, ICP fine alignment, and distance-based heatmap visualization. (PR #7416)
 -   Python 3.13+3.14 support
 
 

examples/python/geometry/point_cloud_comparison.py

@@ -0,0 +1,97 @@
+import open3d as o3d
+import numpy as np
+import matplotlib.pyplot as plt
+import argparse
+
+def preprocess_point_cloud(pcd, voxel_size):
+    """
+    Downsamples and computes FPFH features for global registration.
+    """
+    pcd_down = pcd.voxel_down_sample(voxel_size)
+    radius_normal = voxel_size * 2
+    pcd_down.estimate_normals(
+        o3d.geometry.KDTreeSearchParamHybrid(radius=radius_normal, max_nn=30))
+
+    radius_feature = voxel_size * 5
+    pcd_fpfh = o3d.pipelines.registration.compute_fpfh_feature(
+        pcd_down,
+        o3d.geometry.KDTreeSearchParamHybrid(radius=radius_feature, max_nn=100))
+    return pcd_down, pcd_fpfh
+
+def execute_global_registration(source_down, target_down, source_fpfh, target_fpfh, voxel_size):
+    """
+    Performs RANSAC-based global registration for initial alignment.
+    """
+    distance_threshold = voxel_size * 1.5
+    result = o3d.pipelines.registration.registration_ransac_based_on_feature_matching(
+        source_down, target_down, source_fpfh, target_fpfh, True,
+        distance_threshold,
+        o3d.pipelines.registration.TransformationEstimationPointToPoint(False),
+        3, [
+            o3d.pipelines.registration.CorrespondenceCheckerBasedOnEdgeLength(0.9),
+            o3d.pipelines.registration.CorrespondenceCheckerBasedOnDistance(distance_threshold)
+        ], o3d.pipelines.registration.RANSACConvergenceCriteria(100000, 0.999))
+    return result
+
+def run_comparison(source_path, target_path, is_target_mesh=False):
+    """
+    Complete pipeline: Load -> Global Align -> Fine Align -> Heatmap.
+    """
+    # 1. Data Loading
+    print(":: Loading datasets...")
+    if is_target_mesh:
+        mesh = o3d.io.read_triangle_mesh(target_path)
+        mesh.compute_vertex_normals()
+        # Sampling from mesh to allow distance computation
+        target = mesh.sample_points_poisson_disk(number_of_points=50000)
+    else:
+        target = o3d.io.read_point_cloud(target_path)
+
+    source = o3d.io.read_point_cloud(source_path)
+
+    # 2. Pre-processing & Global Registration
+    voxel_size = 0.05
+    print(":: Performing Global Registration (RANSAC)...")
+    source_down, source_fpfh = preprocess_point_cloud(source, voxel_size)
+    target_down, target_fpfh = preprocess_point_cloud(target, voxel_size)
+
+    global_result = execute_global_registration(source_down, target_down,
+                                               source_fpfh, target_fpfh, voxel_size)
+    source.transform(global_result.transformation)
+
+    # 3. Fine Registration (Point-to-Plane ICP)
+    print(":: Performing Fine Registration (ICP)...")
+    target.estimate_normals()
+    source.estimate_normals()
+    icp_result = o3d.pipelines.registration.registration_icp(
+        source, target, voxel_size, np.identity(4),
+        o3d.pipelines.registration.TransformationEstimationPointToPlane())
+    source.transform(icp_result.transformation)
+
+    # 4. Heatmap Computation
+    distances = source.compute_point_cloud_distance(target)
+    dist_array = np.asarray(distances)
+
+    # Dynamic coloring based on mean distance
+    max_dist = dist_array.mean() * 2
+    colors = plt.get_cmap("jet")(dist_array / max_dist)[:, :3]
+    source.colors = o3d.utility.Vector3dVector(colors)
+
+    print(f":: Final Mean Distance: {np.mean(dist_array):.6f}")
+    o3d.visualization.draw_geometries([source], window_name="Aligned Heatmap Result")
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Professional Point Cloud to Mesh/Cloud Comparison Tool")
+    parser.add_argument("--source", type=str, help="Path to source file")
+    parser.add_argument("--target", type=str, help="Path to target file")
+    parser.add_argument("--is_mesh", action="store_true", help="Flag if target is a Mesh")
+    args = parser.parse_args()
+
+    if args.source and args.target:
+        run_comparison(args.source, args.target, args.is_mesh)
+    else:
+        print(":: No paths provided. Running official Demo data...")
+        knot = o3d.data.KnotMesh()
+        demo_pcd = o3d.data.DemoICPPointClouds()
+        run_comparison(demo_pcd.paths[0], knot.path, is_target_mesh=True)
+