diff --git a/plot_convergence.py b/plot_convergence.py
index fa03105..e398106 100644
--- a/plot_convergence.py
+++ b/plot_convergence.py
@@ -221,7 +221,7 @@ def compare_trainings(list_of_paths, list_of_labels=None):
 
         print(path)
 
-        if ('val_IoUs.txt' in [f for f in listdir(path)]) or ('val_confs.txt' in [f for f in listdir(path)]):
+        if ('val_IoUs.txt' in [f.decode('ascii') for f in listdir(path)]) or ('val_confs.txt' in [f.decode('ascii') for f in listdir(path)]):
             config = Config()
             config.load(path)
         else:
@@ -698,7 +698,7 @@ def experiment_name_1():
 
     # Using the dates of the logs, you can easily gather consecutive ones. All logs should be of the same dataset.
     start = 'Log_2020-04-22_11-52-58'
-    end = 'Log_2020-05-22_11-52-58'
+    end = 'Log_2023-07-29_12-40-27'
 
     # Name of the result path
     res_path = 'results'
@@ -707,9 +707,7 @@ def experiment_name_1():
     logs = np.sort([join(res_path, l) for l in listdir(res_path) if start <= l <= end])
 
     # Give names to the logs (for plot legends)
-    logs_names = ['name_log_1',
-                  'name_log_2',
-                  'name_log_3']
+    logs_names = ['name_log_1']
 
     # safe check log names
     logs_names = np.array(logs_names[:len(logs)])
diff --git a/train_S3DIS.py b/train_S3DIS.py
index 7f07159..7a40612 100644
--- a/train_S3DIS.py
+++ b/train_S3DIS.py
@@ -94,8 +94,10 @@ class S3DISConfig(Config):
                     'resnetb',
                     'resnetb_strided',
                     'resnetb',
+                    'resnetb',
                     'resnetb_strided',
                     'resnetb',
+                    'resnetb',
                     'resnetb_strided',
                     'resnetb',
                     'resnetb',
@@ -119,7 +121,7 @@ class S3DISConfig(Config):
     num_kernel_points = 15
 
     # Radius of the input sphere (decrease value to reduce memory cost)
-    in_radius = 1.2
+    in_radius = 1.8
 
     # Size of the first subsampling grid in meter (increase value to reduce memory cost)
     first_subsampling_dl = 0.03