How to Use a Trained Model for Segmentation of PLY Files?

hello Dr. Gao, Thank you very much for your work. I want to use your trained model to perform segmentation testing on the data, but it didn't run properly. Below is my configuration file and data placement. Could you please tell me where I went wrong? @WeixiaoGao

//** Operation mode **// //--- all modes: Pipeline, Mesh_feature_extraction, Train_config, Test_config, Train_and_Test_config, Data_evaluation_for_all_tiles_config, Save_mesh_features_for_visualization, Class_statistics, Generate_semantic_sampled_points, Moha_Verdi_SOTA_pipeline, Evaluation_SOTA ---//

operating_mode = Test_config

//** Common parameters **// //--- path of data ---//---------------------------------------------------------------------------------------------------------- root_path = M:/umat/data_demo/

//--- path of 'seg_aug.py' ---// seg_aug_py_path = M:/umat/SUMS_Win10_1.3.1/

//--- 0: RF(SUM Paper), 1: SOTA (Competition methods) ---// processing_mode = 0

//--- Select data to process, make sure your data is in the corresponding folder ---// process_train = false process_test = true process_predict = true process_validate = false

//--- Use binary or ascii output ---// use_binary = true

//--- Label definition name ---// label_definition = label

//--- Class name, in *.ply file, make sure that unlabelled is (-1), unclassified is (0), other classes start from (1) ---// labels_name = terrain,high_vegetation,building,water,car,boat

//--- Class color is normalized within [0, 1] ---// labels_color = 170,85,0 ; 0,255,0 ; 255,255,0 ; 0,255,255 ; 255,0,255 ; 0,0,153

//--- labels are ignored in classification, default has no ignored labels --- ignored_labels_name = default

//** Mesh_feature_extraction **// //--- The mesh should have texture, otherwise the classification will not perform very well ---// with_texture = true

//--- Select intermediate data to save ---// save_sampled_pointclouds = false save_oversegmentation_mesh = true save_tex_cloud = false

//--- For compute relative elevation (meters) ---// multi_scale_ele_radius = 10.0,20.0,40.0 long_range_radius_default = default local_ground_segs = default

//--- For generate HSV histogram ---// hsv_bins = 15,5,5

//--- Interior medial axis transform parameters ---// mat_delta_convergance = default mat_initialized_radius = default mat_denoising_seperation_angle = default mat_iteration_limit_number = default

//--- For custom or existing segments, make sure you have the field face_segment_id ---// use_existing_mesh_segments_on_training = false use_existing_mesh_segments_on_testing = true use_existing_mesh_segments_on_predicting = true use_existing_mesh_segments_on_validation = false

//--- settings for texture color processing --- // //If it is set to true then it consumes a lot of memory, if not then we use face average color which save memories but it has less accurate color feature //recommendation: adaptive triangles: true, dense triangles: false; use_face_pixels_color_aggregation_on_training = true use_face_pixels_color_aggregation_on_testing = true use_face_pixels_color_aggregation_on_predicting = true use_face_pixels_color_aggregation_on_validation = false

//--- Region growing for mesh over-segmentation ---// mesh_distance_to_plane = 0.5 mesh_accepted_angle = 90.0 mesh_minimum_region_size = 0.0 partition_folder_path = segments/ partition_prefixs = _mesh_seg

//--- Feature selection, check the end of this file for feature dictionary; the multiscale elevation (use_mulsc_eles) corresponding to (multi_scale_ele_radius) ---// use_feas = 0,1,2,3 use_mulsc_eles = 0,1,2 use_basic_features = 0,1,2,3,4 use_eigen_features = 3,4,6,11 use_color_features = 3,4,5,6,10,11,12,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38

//--- Used sampled point clouds instead of triangle faces (only for merged mesh that has topological interstice); In order to compute relative elevation faster, a sparse point cloud will be sampled by default, these settings are for all datasets (train, test, predict, validate) ---// is_pointclouds_exist = false sampling_point_density = 1.0 ele_sampling_point_density = default

//---- Use which strategy to sample point clouds on mesh ---// //0: sampled only; 1: face center only; 2: face center plus face vertices; 3: sampled and face center; 4: sampled and face center plus face vertices. //0: for relative elevation (automatic); 1,2: for mesh feature computation; 3,4: for merged mesh feature computation; 3: for deep learning sampled point clouds.
sampling_strategy_training = 3 sampling_strategy_testing = 2 sampling_strategy_predicting = 2 sampling_strategy_validation = 2

//** Train_and_Test parameters **// //--- Random forest parameters ---// rf_tree_numbers = 200 rf_tree_depth = 50

//--- smote data augmentation, it will call python file, make sure you have installed 'imbalanced-learn' and replaced 'filter.py' in 'Lib\site-packages\imblearn\over_sampling_smote' with ours ---// augmented_data_exist = false enable_augment = false used_k_neighbors = 15

//--- save intermediate data when run the 'Test' ---// save_feature_importance = true save_error_map = true

//** Class_statistics **// //--- 0: mesh, 1: sampled point cloud ---// input_type_for_statistics = 0

//** SOTA: Moha_Verdi_SOTA_pipeline **// //--- Mesh neighborhood radius, for initial planarity to generate region growing seeds ---// short_range_radius_default = default

//--- Mesh neighborhood radius, for face color ---// mr_facet_neg_sphericial_radius = default

//--- Mesh region growing parameters ---// mr_limit_distance = default mr_angle_thres = default mr_L1_color_dis = default mr_max_sp_area = default

//--- MRF formulation, regularization parameter is (mrf_lambda_mh) ---// mrf_lambda_mh = 0.5 mrf_energy_amplify = default

//** SOTA : all deep learning methods **// //--- add point color for sampled point cloud, tune 'sampling_point_density' for sampling density ---// add_point_color_for_dp_input = true

//--- label prefix of output *.ply from competitive deep learning method ---// //--- SPG:"pred", KPConv: "preds", RandLanet: "label", PointNet, PointNet2: "pred" ---// label_string = preds

//--- 0: no minus, if label start from 0; 1: minus 1, if label start from 1 ---// //---SPG: 1; KPConv: 0; RandLanet: 1; PointNet2: 0 ; PointNet: 0 ---// label_minus = 0

//--- Equal to the original sampled point cloud or not ---// //--- others: true, Randlanet: false ---// equal_cloud = true

//** Batch processing parameters **// //--- If the 'batchnames**.txt' exists then true ---// use_existing_splited_batch = false

//--- Merge small tiles into one mesh (batch_size=row, sub_batch_size=column)---// batch_size = 1 sub_batch_size = 1

//--- Check which data use batch processing and point cloud sampling ---// use_batch_processing_on_training = false use_batch_processing_on_testing = false use_batch_processing_on_predicting = false use_batch_processing_on_validation = false

//--- Use point cloud region growing or use mesh region growing --- use_pointcloud_region_growing_on_training = false use_pointcloud_region_growing_on_testing = false use_pointcloud_region_growing_on_predicting = false use_pointcloud_region_growing_on_validation = false

//--- Merge segments, usually for merged mesh region growing ---// use_merged_segments_on_training = false use_merged_segments_on_testing = false use_merged_segments_on_predicting = false use_merged_segments_on_validation = false adjacent_radius = 0.5 adjacent_pt2plane_distance = 0.25 adjacent_seg_angle = 90

//--- Point cloud region growing parameters ---// //adpative mesh: 0.5, 90; dense mesh: 0.3, 45 pcl_distance_to_plane = 0.5 pcl_accepted_angle = 45 pcl_minimum_region_size = 0 pcl_k_nn = 12

//---save texture in the output folder of each batch from original input ---// save_textures_in_predict = false

//** Feature dictionary specification (not parameters) **// use_feas { {segment_basic_features, 0}, {segment_eigen_features, 1}, {segment_color_features, 2}, {elevation_features, 3} };

use_mulsc_eles { {"10.0", 0}, {"20.0", 1}, {"40.0", 2} };

basic_feature_base_names { {"avg_center_z", 0}, {"interior_mat_radius", 1}, {"sum_area", 2}, {"relative_elevation", 3}, {"triangle_density", 4} };

eigen_feature_base_names { {"eigen_1", 0}, {"eigen_2", 1}, {"eigen_3", 2}, {"verticality", 3}, {"linearity", 4}, {"planarity", 5}, {"sphericity", 6}, {"anisotropy", 7}, {"eigenentropy", 8}, {"omnivariance", 9}, {"sumeigenvals", 10}, {"curvature", 11}, {"verticality_eig1", 12}, {"verticality_eig3", 13}, {"surface", 14}, {"volume", 15}, {"absolute_eigvec_1_moment_1st", 16}, {"absolute_eigvec_2_moment_1st", 17}, {"absolute_eigvec_3_moment_1st", 18}, {"absolute_eigvec_1_moment_2nd", 19}, {"absolute_eigvec_2_moment_2nd", 20}, {"absolute_eigvec_3_moment_2nd", 21}, {"vertical_moment_1st", 22}, {"vertical_moment_2nd", 23}, {"uniformity", 24} };

color_feature_base_names { {"red", 0}, {"green", 1}, {"blue", 2}, {"hue", 3}, {"sat", 4}, {"val", 5}, {"greenness", 6}, {"red_var", 7}, {"green_var", 8}, {"blue_var", 9}, {"hue_var", 10}, {"sat_var", 11}, {"val_var", 12}, {"greenness_var", 13}, {"hue_bin_0", 14}, {"hue_bin_1", 15}, {"hue_bin_2", 16}, {"hue_bin_3", 17}, {"hue_bin_4", 18}, {"hue_bin_5", 19}, {"hue_bin_6", 20}, {"hue_bin_7", 21}, {"hue_bin_8", 22}, {"hue_bin_9", 23}, {"hue_bin_10", 24}, {"hue_bin_11", 25}, {"hue_bin_12", 26}, {"hue_bin_13", 27}, {"hue_bin_14", 28}, {"sat_bin_0", 29}, {"sat_bin_1", 30}, {"sat_bin_2", 31}, {"sat_bin_3", 32}, {"sat_bin_4", 33}, {"val_bin_0", 34}, {"val_bin_1", 35}, {"val_bin_2", 36}, {"val_bin_3", 37}, {"val_bin_4", 38} };

the error message as following: (sums) PS M:\umat\SUMS_Win10_1.3.1> ./semantic_urban_mesh_segmentation.exe M:/umat/data_demo/

Reading configuration file: Done in (s): 0.0002698

--------------------- Testing Mode --------------------- Get all test data. Done in (s): 0.0002158

Loading trained model -> Done ! Start to read features Tile+1976+2706 RPly: Unable to open file failed to open ply file 'M:/umat/datademo/feature/test/Tile+1976_+2706_pcl_feas.ply' reading file failed (no data exist) File loading failed

if set "operating_mode = Mesh_feature_extraction", the error:

(sums) PS M:\umat\SUMS_Win10_1.3.1> ./semantic_urban_mesh_segmentation.exe M:/umat/data_demo/

Reading configuration file: Done in (s): 0.0001747

--------------------- Generating test mesh features --------------------- Get all test data. Done in (s): 0.0001309

loading existing mesh segments: Tile+1976+2706 RPly: Unable to open file failed to open ply file 'M:/umat/datademo/segments/test/Tile+1976_+2706_mesh_seg.ply' reading file failed (no data exist) File loading failed (sums) PS M:\umat\SUMS_Win10_1.3.1>

others:

operating_mode = Pipeline

//** Common parameters **// //--- path of data ---//---------------------------------------------------------------------------------------------------------- root_path = M:/umat/data_demo/

//--- path of 'seg_aug.py' ---// seg_aug_py_path = M:/umat/SUMS_Win10_1.3.1/

//--- 0: RF(SUM Paper), 1: SOTA (Competition methods) ---// processing_mode = 0

//--- Select data to process, make sure your data is in the corresponding folder ---// process_train = true process_test = true process_predict = true process_validate = true

//--- Use binary or ascii output ---// use_binary = false

//--- Label definition name ---// label_definition = label

//--- Class name, in *.ply file, make sure that unlabelled is (-1), unclassified is (0), other classes start from (1) ---// labels_name = terrain,high_vegetation,building,water,car,boat

//--- Class color is normalized within [0, 1] ---// labels_color = 170,85,0 ; 0,255,0 ; 255,255,0 ; 0,255,255 ; 255,0,255 ; 0,0,153

//--- labels are ignored in classification, default has no ignored labels --- ignored_labels_name = default

//** Mesh_feature_extraction **// //--- The mesh should have texture, otherwise the classification will not perform very well ---// with_texture = true

//--- Select intermediate data to save ---// save_sampled_pointclouds = false save_oversegmentation_mesh = true save_tex_cloud = false

//--- For compute relative elevation (meters) ---// multi_scale_ele_radius = 10.0,20.0,40.0 long_range_radius_default = default local_ground_segs = default

//--- For generate HSV histogram ---// hsv_bins = 15,5,5

//--- Interior medial axis transform parameters ---// mat_delta_convergance = default mat_initialized_radius = default mat_denoising_seperation_angle = default mat_iteration_limit_number = default

//--- For custom or existing segments, make sure you have the field face_segment_id ---// use_existing_mesh_segments_on_training = false use_existing_mesh_segments_on_testing = false use_existing_mesh_segments_on_predicting = false use_existing_mesh_segments_on_validation = false