From 79605dff3850bb58b2c55d31e96deb1987e16b14 Mon Sep 17 00:00:00 2001
From: Maciej Wielgosz <maciej.wielgosz@nibio.no>
Date: Tue, 3 Jan 2023 12:40:26 +0100
Subject: [PATCH] instance segmentation metrics visualizer
---
...instance_segmentation_metrics_in_folder.py | 2 +-
visualization/inst_seg_visualizer.py | 122 ++++++++++++++++++
2 files changed, 123 insertions(+), 1 deletion(-)
create mode 100644 visualization/inst_seg_visualizer.py
diff --git a/metrics/instance_segmentation_metrics_in_folder.py b/metrics/instance_segmentation_metrics_in_folder.py
index 9bc000a..8ddd37b 100644
--- a/metrics/instance_segmentation_metrics_in_folder.py
+++ b/metrics/instance_segmentation_metrics_in_folder.py
@@ -160,7 +160,7 @@ class InstanceSegmentationMetricsInFolder():
AttachLabelsToLasFile(
gt_las_file_path,
target_las_file_path,
- update_las_file_path = os.path.join(self.output_folder_path, gt_las_file_core_name + '_updated.las'),
+ update_las_file_path = os.path.join(self.output_folder_path, gt_las_file_core_name + '.las'),
gt_label_name='treeID',
target_label_name='treeID',
verbose=self.verbose
diff --git a/visualization/inst_seg_visualizer.py b/visualization/inst_seg_visualizer.py
new file mode 100644
index 0000000..cdf6fa3
--- /dev/null
+++ b/visualization/inst_seg_visualizer.py
@@ -0,0 +1,122 @@
+import argparse
+import glob
+import os
+import laspy
+import numpy as np
+import pandas as pd
+from tqdm import tqdm
+
+class InstSegVisualizer:
+ GT_LABEL_NAME = 'treeID'
+ TARGET_LABEL_NAME = 'instance_nr'
+ CSV_GT_LABEL_NAME = 'gt_label(dominant_label)'
+ CSV_TARGET_LABEL_NAME = 'pred_label'
+
+ def __init__(self, folder_with_metrics, verbose) -> None:
+ self.folder_with_metrics = folder_with_metrics
+ self.verbose = verbose
+
+ def get_las_file_paths(self, folder_path):
+ las_file_paths = glob.glob(folder_path + '/*.las', recursive=False)
+ las_file_paths.sort()
+ return las_file_paths
+
+ def get_csv_file_paths(self, folder_path):
+ csv_file_paths = glob.glob(folder_path + '/*.csv', recursive=False)
+ csv_file_paths.sort()
+ return csv_file_paths
+
+ def match_las_and_csv_files(self, las_file_paths, csv_file_paths):
+ matched_paths = []
+ for las_file_path, csv_file_path in zip(las_file_paths, csv_file_paths):
+ las_file_name = las_file_path.split('/')[-1]
+ csv_file_name = csv_file_path.split('/')[-1]
+ if las_file_name.split('.')[0] == csv_file_name.split('.')[0]:
+ matched_paths.append((las_file_path, csv_file_path))
+ else:
+ raise Exception('The las file name and the csv file name do not match')
+ return matched_paths
+
+ def get_gt_and_pred_labels_from_csv_for_single_file(self, csv_file_path):
+ df = pd.read_csv(csv_file_path)
+ gt_labels = df[self.CSV_GT_LABEL_NAME].values
+ pred_labels = df[self.CSV_TARGET_LABEL_NAME].values
+ # map to int
+ gt_labels = gt_labels.astype(int)
+ pred_labels = pred_labels.astype(int)
+ # create a list of tuples
+ matched_labels = list(zip(gt_labels, pred_labels))
+ return matched_labels
+
+ def extract_overlapping_pc_in_single_file(self, las_file_path, matched_labels):
+ # get the las file name
+ las_file_name = las_file_path.split('/')[-1]
+
+ # create if does not exist the folder with metrics named after the las file
+ if not os.path.exists(self.folder_with_metrics + '/' + las_file_name.split('.')[0]):
+ os.makedirs(self.folder_with_metrics + '/' + las_file_name.split('.')[0])
+
+ for gt_label, pred_label in tqdm(matched_labels):
+ # read the las file
+ las_file = laspy.read(las_file_path)
+
+ # get GT_LABEL_NAME points
+ gt_points = las_file.points[las_file.points[self.GT_LABEL_NAME] == gt_label]
+ # create a new array of gt points with the same shape as the red array and fill it with 255
+ gt_points['red'] = np.ones(gt_points['red'].shape) * 255
+ gt_points['green'] = np.ones(gt_points['green'].shape) * 0
+ gt_points['blue'] = np.ones(gt_points['blue'].shape) * 0
+
+ # get TARGET_LABEL_NAME points
+ pred_points = las_file.points[las_file.points[self.TARGET_LABEL_NAME] == pred_label]
+
+ # create a new array of pred points with the same shape as the blue array and fill it with 255
+ pred_points['red'] = np.ones(pred_points['red'].shape) * 0
+ pred_points['green'] = np.ones(pred_points['green'].shape) * 0
+ pred_points['blue'] = np.ones(pred_points['blue'].shape) * 255
+
+ # dump both arrays to las to the folder with metrics named after the las file
+ las_file.points = gt_points
+ las_file.write(self.folder_with_metrics + '/' + las_file_name.split('.')[0] + '/' + las_file_name.split('.')[0] + '_gt_' + str(gt_label) + '.las')
+ las_file.points = pred_points
+ las_file.write(self.folder_with_metrics + '/' + las_file_name.split('.')[0] + '/' + las_file_name.split('.')[0] + '_pred_' + str(pred_label) + '.las')
+
+ # merge using pdal and os.system and save the merged file in the folder with metrics named after the las file
+ os.system(
+ 'pdal merge ' + self.folder_with_metrics + '/' +
+ las_file_name.split('.')[0] + '/' + las_file_name.split('.')[0] +
+ '_gt_' + str(gt_label) + '.las ' + self.folder_with_metrics + '/' +
+ las_file_name.split('.')[0] + '/' + las_file_name.split('.')[0] +
+ '_pred_' + str(pred_label) + '.las ' + self.folder_with_metrics + '/' +
+ las_file_name.split('.')[0] + '/' + las_file_name.split('.')[0] +
+ '_gt_' + str(gt_label) + '_pred_' + str(pred_label) + '.las'
+ )
+
+ def extract_overlapping_pc_of_all_files(self, matched_files):
+ for las_file_path, csv_file_path in matched_files:
+ if self.verbose:
+ print('Processing file: ', las_file_path)
+ matched_labels = self.get_gt_and_pred_labels_from_csv_for_single_file(csv_file_path)
+ self.extract_overlapping_pc_in_single_file(las_file_path, matched_labels)
+
+ def main(self):
+ las_file_paths = self.get_las_file_paths(self.folder_with_metrics)
+ csv_file_paths = self.get_csv_file_paths(self.folder_with_metrics)
+ matched_files = self.match_las_and_csv_files(las_file_paths, csv_file_paths)
+ self.extract_overlapping_pc_of_all_files(matched_files)
+
+ if self.verbose:
+ # print the number of las files
+ print('Number of las files: ', len(las_file_paths))
+ # print where the files were saved
+ print('The files were saved in: ', self.folder_with_metrics)
+
+if __name__ == '__main__':
+ # use argparse to get the folder with metrics
+ parser = argparse.ArgumentParser()
+ parser.add_argument('--folder_with_metrics', type=str, default='metrics')
+ parser.add_argument('--verbose', help="Print more information.", action="store_true")
+ args = parser.parse_args()
+
+ InstSegVisualizer(args.folder_with_metrics, args.verbose).main()
+
--
GitLab