update codes

Author: JokerJohn

README.md

@@ -27,6 +27,7 @@ These complementary aspects require different evaluation approaches, as global d
 
 ## News
 
+- **2025/07/05:** Update the parameter setting and more config examples.
 - **2025/05/05:** Add new test data and remove the simulation codes.
 - **2025/03/05**: [Formally published](https://ieeexplore.ieee.org/document/10910156)!
 - **2025/02/25**: Accept!

map_eval/.idea/libraries/ROS.xml

@@ -4,6 +4,9 @@ project(map_eval)
 set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_BUILD_TYPE Release)
 
+set(Open3D_DIR "/home/xchu/xicp_ws/devel/.private/open3d_catkin/lib/cmake/Open3D")
+message(STATUS "LTLoc: Set Open3D_DIR to ${Open3D_DIR}") # 打印信息，确认设置生效
+
 find_package(Open3D REQUIRED)
 find_package(Eigen3 REQUIRED)
 find_package(OpenMP REQUIRED)

map_eval/CMakeLists.txt

@@ -57,4 +57,13 @@ evaluate_noise_gt: false
 
 # vmd voxel size
 # outdoor: 2.0-4.0; indoor: 2.0-3.0
-vmd_voxel_size: 3.0
+vmd_voxel_size: 3.0
+
+# downsample size for estimated and gt point cloud map
+downsample_size: 0.01
+
+# if show the evaluation error map after registration
+use_visualization: true
+
+# if print the debug info
+enable_debug: true

map_eval/config/config.yaml

@@ -0,0 +1,69 @@
+# registration methods from Open3D: 0-point-to-point icp; 1-point-to-plane icp; 2-GICP
+registration_methods: 2
+# set icp_max_distance when search the nearest point, larger value will make the search slower
+icp_max_distance: 1.0
+
+# accuracy_level, vector5d, we mainly use the result of the first element
+# if inlier is very small, we can try to larger the value, e.g. for outdoors, [0.5, 0.3, 0.2, 0.1, 0.05]
+accuracy_level: [0.2, 0.1, 0.08, 0.05, 0.01]
+
+# initial_matrix, vector16d, the initial matrix of the registration
+# make sure the format is correct, or you will got the error log: YAML::BadSubscript' what():  operator[] call on a scalar
+# if you evaluate the LIO mapping result, you can use the CloudCompare to get the initial matrix first
+initial_matrix:
+  - [1.0, 0.0, 0.0, 0.0]
+  - [0.0, 1.0, 0.0, 0.0]
+  - [0.0, 0.0, 1.0, 0.0]
+  - [0.0, 0.0, 0.0, 1.0]
+
+# estimate map folder path, end with "/", make sure your estimate point cloud and map is renamed as "map.pcd"
+# results path will also in the same folder:  estimate_map_path + "map_results"
+estimate_map_path: /home/xchu/data/ltloc_result/building_day_all/20220225_building_day_ref_tsvd/
+
+# ground truth map file path
+# note that we support ".pcd" or ".ply" format for ground truth map loading
+gt_map_path: /media/xchu/e81eaf80-d92c-413a-a503-1c9b35b19963/home/xchu/data/evaluation/gt/map/building_day_gt.pcd
+
+
+# scene name, not important, just for batch processing for a series of scenes
+scene_name: stairs
+
+# other settings
+# if we save the immediate result to a text file
+save_immediate_result: true
+
+# if evaluate Mean Map Entropy (MME) if you do not have ground truth map
+# but remember, MME can be only used for map comparison with the same scale, e.g., maps are for odometry.
+# it costs a lot of time to calculate MME, so we set it as false by default
+evaluate_mme: false
+
+# if we have ground truth map, we also want to evaluate the MME of the ground truth map
+evaluate_gt_mme: false
+
+# NN search radius for MME calculation (m), larger value will make the search much more slower, even the memory overflow
+nn_radius: 0.1
+
+# if we want to evaluate the registration result using initial matrix without alignments, this operation will save time
+# that means your estimate map is already aligned with the ground truth map, 
+# or your estimate map and the ground truth map is already in the same coordinate system
+# e.g. the estimate map is already aligned with the ground truth map
+evaluate_using_initial: false
+
+# we want to add noise to the ground truth map, and evaluate the registration result
+# note that true will cause the system do not load the estimate map, instead the noise gt_map will be used
+# just for algorithm robustness evaluation experiments, not for real-world application
+# do not recommand to set it as true, since it will cause the memory overflow
+evaluate_noise_gt: false
+
+# vmd voxel size
+# outdoor: 2.0-4.0; indoor: 2.0-3.0
+vmd_voxel_size: 2.0
+
+# downsample size for estimated and gt point cloud map
+downsample_size: 0.01
+
+# if show the evaluation error map after registration
+use_visualization: true
+
+# if print the debug info
+enable_debug: true

map_eval/config/config_building_day.yaml

@@ -0,0 +1,69 @@
+# registration methods from Open3D: 0-point-to-point icp; 1-point-to-plane icp; 2-GICP
+registration_methods: 2
+# set icp_max_distance when search the nearest point, larger value will make the search slower
+icp_max_distance: 1.0
+
+# accuracy_level, vector5d, we mainly use the result of the first element
+# if inlier is very small, we can try to larger the value, e.g. for outdoors, [0.5, 0.3, 0.2, 0.1, 0.05]
+accuracy_level: [0.2, 0.1, 0.08, 0.05, 0.01]
+
+# initial_matrix, vector16d, the initial matrix of the registration
+# make sure the format is correct, or you will got the error log: YAML::BadSubscript' what():  operator[] call on a scalar
+# if you evaluate the LIO mapping result, you can use the CloudCompare to get the initial matrix first
+initial_matrix:
+  - [1.0, 0.0, 0.0, 0.0]
+  - [0.0, 1.0, 0.0, 0.0]
+  - [0.0, 0.0, 1.0, 0.0]
+  - [0.0, 0.0, 0.0, 1.0]
+
+# estimate map folder path, end with "/", make sure your estimate point cloud and map is renamed as "map.pcd"
+# results path will also in the same folder:  estimate_map_path + "map_results"
+estimate_map_path: /home/xchu/data/ltloc_result/corridor_all/20220216_corridor_day_ref_eular/
+
+# ground truth map file path
+# note that we support ".pcd" or ".ply" format for ground truth map loading
+gt_map_path: /media/xchu/e81eaf80-d92c-413a-a503-1c9b35b19963/home/xchu/data/evaluation/gt/map/corridor_day_gt.pcd
+
+
+# scene name, not important, just for batch processing for a series of scenes
+scene_name: corridor
+
+# other settings
+# if we save the immediate result to a text file
+save_immediate_result: true
+
+# if evaluate Mean Map Entropy (MME) if you do not have ground truth map
+# but remember, MME can be only used for map comparison with the same scale, e.g., maps are for odometry.
+# it costs a lot of time to calculate MME, so we set it as false by default
+evaluate_mme: false
+
+# if we have ground truth map, we also want to evaluate the MME of the ground truth map
+evaluate_gt_mme: false
+
+# NN search radius for MME calculation (m), larger value will make the search much more slower, even the memory overflow
+nn_radius: 0.1
+
+# if we want to evaluate the registration result using initial matrix without alignments, this operation will save time
+# that means your estimate map is already aligned with the ground truth map, 
+# or your estimate map and the ground truth map is already in the same coordinate system
+# e.g. the estimate map is already aligned with the ground truth map
+evaluate_using_initial: false
+
+# we want to add noise to the ground truth map, and evaluate the registration result
+# note that true will cause the system do not load the estimate map, instead the noise gt_map will be used
+# just for algorithm robustness evaluation experiments, not for real-world application
+# do not recommand to set it as true, since it will cause the memory overflow
+evaluate_noise_gt: false
+
+# vmd voxel size
+# outdoor: 2.0-4.0; indoor: 2.0-3.0
+vmd_voxel_size: 2.0
+
+# downsample size for estimated and gt point cloud map
+downsample_size: 0.01
+
+# if show the evaluation error map after registration
+use_visualization: true
+
+# if print the debug info
+enable_debug: true

map_eval/config/config_corridor.yaml

@@ -0,0 +1,69 @@
+# registration methods from Open3D: 0-point-to-point icp; 1-point-to-plane icp; 2-GICP
+registration_methods: 2
+# set icp_max_distance when search the nearest point, larger value will make the search slower
+icp_max_distance: 1.0
+
+# accuracy_level, vector5d, we mainly use the result of the first element
+# if inlier is very small, we can try to larger the value, e.g. for outdoors, [0.5, 0.3, 0.2, 0.1, 0.05]
+accuracy_level: [0.2, 0.1, 0.08, 0.05, 0.01]
+
+# initial_matrix, vector16d, the initial matrix of the registration
+# make sure the format is correct, or you will got the error log: YAML::BadSubscript' what():  operator[] call on a scalar
+# if you evaluate the LIO mapping result, you can use the CloudCompare to get the initial matrix first
+initial_matrix:
+  - [1.0, 0.0, 0.0, 0.0]
+  - [0.0, 1.0, 0.0, 0.0]
+  - [0.0, 0.0, 1.0, 0.0]
+  - [0.0, 0.0, 0.0, 1.0]
+
+# estimate map folder path, end with "/", make sure your estimate point cloud and map is renamed as "map.pcd"
+# results path will also in the same folder:  estimate_map_path + "map_results"
+estimate_map_path: /home/xchu/data/ltloc_result/stair_bob_all/stairs_bob_fcn_sr/
+
+# ground truth map file path
+# note that we support ".pcd" or ".ply" format for ground truth map loading
+gt_map_path: /media/xchu/新加卷/GEODE/gt_map.pcd
+
+
+# scene name, not important, just for batch processing for a series of scenes
+scene_name: stairs
+
+# other settings
+# if we save the immediate result to a text file
+save_immediate_result: true
+
+# if evaluate Mean Map Entropy (MME) if you do not have ground truth map
+# but remember, MME can be only used for map comparison with the same scale, e.g., maps are for odometry.
+# it costs a lot of time to calculate MME, so we set it as false by default
+evaluate_mme: false
+
+# if we have ground truth map, we also want to evaluate the MME of the ground truth map
+evaluate_gt_mme: false
+
+# NN search radius for MME calculation (m), larger value will make the search much more slower, even the memory overflow
+nn_radius: 0.1
+
+# if we want to evaluate the registration result using initial matrix without alignments, this operation will save time
+# that means your estimate map is already aligned with the ground truth map, 
+# or your estimate map and the ground truth map is already in the same coordinate system
+# e.g. the estimate map is already aligned with the ground truth map
+evaluate_using_initial: false
+
+# we want to add noise to the ground truth map, and evaluate the registration result
+# note that true will cause the system do not load the estimate map, instead the noise gt_map will be used
+# just for algorithm robustness evaluation experiments, not for real-world application
+# do not recommand to set it as true, since it will cause the memory overflow
+evaluate_noise_gt: false
+
+# vmd voxel size
+# outdoor: 2.0-4.0; indoor: 2.0-3.0
+vmd_voxel_size: 2.0
+
+# downsample size for estimated and gt point cloud map
+downsample_size: 0.01
+
+# if show the evaluation error map after registration
+use_visualization: true
+
+# if print the debug info
+enable_debug: true