add of fusion support for Semantic8 and NPM3D

examples/npm3d/README.md

@@ -33,4 +33,19 @@ python npm3d_seg.py --rootdir path_to_data_dir --savedir path_to_save_dir --test
 If the model was trained without using the lidar intensity:
 ```
 python npm3d_seg.py --rootdir path_to_data_dir --savedir path_to_save_dir --test --nocolor
+```
+
+**note**: the `test_step` parameter is set `0.8`. It is possible to change it. A smaller step of sliding window would produce better segmentation at a the cost of a longer computation time.
+
+## Fusion model
+
+Once models (RGB and without color information) have been trained, it is possible to train a fusion model.
+
+### Training
+```
+python npm33d_seg_fusion.py --rootdir path_to_data_processed --savedir path_to_save_dirctory --model_rgb path_to_rgb_model_directory --model_noc path_to_no_color_model_directory
+```
+### Test
+```
+python npm3d_seg_fusion.py --rootdir path_to_data_processeed --savedir path_to_save_dirctory --model_rgb path_to_rgb_model_directory --model_noc path_to_no_color_model_directory --test --savepts
 ```

examples/npm3d/npm3d_seg.py

@@ -238,10 +238,8 @@ def main():
 
      # create the filelits (train / val) according to area
     print("Create filelist...", end="")
-    filelist_train = []
     train_dir = os.path.join(args.rootdir, "train_pointclouds")
     filelist_train = [dataset for dataset in os.listdir(train_dir)]
-    filename_test = []
     test_dir = os.path.join(args.rootdir, "test_pointclouds")
     filelist_test = [dataset for dataset in os.listdir(test_dir)]
     print(f"done, {len(filelist_train)} train files, {len(filelist_test)} test files")