diff --git a/main_inference.py b/main_inference.py
new file mode 100644
index 0000000..7b72616
--- /dev/null
+++ b/main_inference.py
@@ -0,0 +1,73 @@
+from pathlib import Path
+
+import torch
+from torch_geometric.data import Data
+from tqdm import tqdm
+
+from torch.utils.data import DataLoader
+
+# Dataset and Dataloaders
+# =============================================================================
+
+# Transforms
+from ml_lib.point_toolset.point_io import BatchToData
+from ml_lib.utils.model_io import SavedLightningModels
+
+
+# Datasets
+from datasets.full_pointclouds import FullCloudsDataset
+from utils.project_config import GlobalVar, ThisConfig
+
+
+def prepare_dataloader(config_obj):
+    dataset = FullCloudsDataset(config_obj.data.root, split=GlobalVar.data_split.test,
+                                setting=GlobalVar.settings[config_obj.model.type])
+    # noinspection PyTypeChecker
+    return DataLoader(dataset, batch_size=config_obj.train.batch_size,
+                      num_workers=config_obj.data.worker, shuffle=False)
+
+
+def restore_logger_and_model(log_dir):
+    model = SavedLightningModels.load_checkpoint(models_root_path=log_dir, n=-1)
+    model = model.restore()
+    if torch.cuda.is_available():
+        model.cuda()
+    else:
+        model.cpu()
+    return model
+
+
+if __name__ == '__main__':
+    outpath = Path('output')
+    model_path = Path('/home/steffen/projects/point_to_primitive/output/P2G/PG_9f7ac027e3359fa5f5e5bcd32044a167/version_69')
+    config_filename = 'config.ini'
+    inference_out = 'manual_test_out.csv'
+
+    config = ThisConfig()
+    config.read_file((Path(model_path) / config_filename).open('r'))
+    test_dataloader = prepare_dataloader(config)
+
+    loaded_model = restore_logger_and_model(model_path)
+    loaded_model.eval()
+
+    with (model_path / inference_out).open(mode='w') as outfile:
+        outfile.write(f'{",".join(FullCloudsDataset.headers[:6])},class,cluster\n')
+        batch_to_data = BatchToData()
+        for batch_pos_x_n_y_c in tqdm(test_dataloader, total=len(test_dataloader)):
+            data = batch_to_data(*batch_pos_x_n_y_c) if not isinstance(batch_pos_x_n_y_c, Data) else batch_pos_x_n_y_c
+            y = loaded_model(data.to(device='cuda' if torch.cuda.is_available() else 'cpu'))
+            y_primary = torch.argmax(y.main_out, dim=-1).squeeze().cpu().numpy()
+            y_sec = -1
+            try:
+                y_sec = torch.argmax(y.grid_out, dim=-1).squeeze().cpu().numpy()
+            except AttributeError:
+                pass
+            try:
+                y_sec = torch.argmax(y.prim_out, dim=-1).squeeze().cpu().numpy()
+            except AttributeError:
+                pass
+            for row in range(data.num_nodes):
+                outfile.write(f'{",".join(map(str, data.pos[row].tolist()))},' +
+                              f'{",".join(map(str, data.x[row].tolist()))},' +
+                              f'{y_primary[row]},{y_sec[row]}\n')
+    print('Done')
diff --git a/utils/project_config.py b/utils/project_config.py
index 5cd0b11..c2c96d1 100644
--- a/utils/project_config.py
+++ b/utils/project_config.py
@@ -17,6 +17,9 @@ class DataClass(Namespace):
     def __repr__(self):
         return f'{self.__class__.__name__}({self.__dict__().__repr__()})'
 
+    def __getitem__(self, item):
+        return self.__getattribute__(item)
+
 
 class Classes(DataClass):
     # Object Classes for Point Segmentation
@@ -29,6 +32,12 @@ class Classes(DataClass):
     Plane = 6
 
 
+class Settings(DataClass):
+    P2G = 'grid'
+    P2P = 'prim'
+    PN2 = 'pc'
+
+
 class DataSplit(DataClass):
     # DATA SPLIT OPTIONS
     train = 'train'
@@ -49,6 +58,8 @@ class GlobalVar(DataClass):
 
     prim_count = -1
 
+    settings = Settings()
+
 
 from models import *