.

doc/pretrained_models_guide.md

@@ -4,15 +4,14 @@
 
 ### Models
 
-We provide pretrained weights for S3DIS dataset. The raw weights come with a parameter file describing the architecture and network hyperparameters. THe code can thus load the network automatically.
+We provide pretrained weights for S3DIS dataset. The raw weights come with a parameter file describing the architecture and network hyperparameters. The code can thus load the network automatically.
-
-The instructions to run these models are in the S3DIS documentation, section [Test the trained model](./doc/scene_segmentation_guide.md#test-the-trained-model).
 
 
 | Name (link) | KPConv Type | Description | Score |
 |:-------------|:-------------:|:-----|:-----:|
 | [Light_KPFCNN](https://drive.google.com/file/d/14sz0hdObzsf_exxInXdOIbnUTe0foOOz/view?usp=sharing) | rigid | A network with small `in_radius` for light GPU consumption (~8GB) | 65.4% |
 | [Heavy_KPFCNN](https://drive.google.com/file/d/1ySQq3SRBgk2Vt5Bvj-0N7jDPi0QTPZiZ/view?usp=sharing) | rigid | A network with better performances but needing bigger GPU (>18GB). | 66.4% |
+| [Deform_KPFCNN](https://drive.google.com/file/d/1ObGr2Srfj0f7Bd3bBbuQzxtjf0ULbpSA/view?usp=sharing) | rigid | Deformable convolution network needing big GPU (>20GB). | 67.3% |
 
 
 

plot_convergence.py

@@ -47,6 +47,22 @@ from datasets.SemanticKitti import SemanticKittiDataset
 #       \***********************/
 #
 
+def listdir_str(path):
+
+    # listdir can return binary string instead od decoded string sometimes.
+    # This function ensures a steady behavior
+
+    f_list = []
+    for f in listdir(path):
+        try:
+            f = f.decode()
+        except (UnicodeDecodeError, AttributeError):
+            pass
+        f_list.append(f)
+
+    return f_list
+
+
 
 def running_mean(signal, n, axis=0, stride=1):
     signal = np.array(signal)
@@ -148,7 +164,7 @@ def load_single_IoU(filename, n_parts):
 
 def load_snap_clouds(path, dataset, only_last=False):
 
-    cloud_folders = np.array([join(path, f) for f in listdir(path) if f.startswith('val_preds')])
+    cloud_folders = np.array([join(path, f) for f in listdir_str(path) if f.startswith('val_preds')])
     cloud_epochs = np.array([int(f.split('_')[-1]) for f in cloud_folders])
     epoch_order = np.argsort(cloud_epochs)
     cloud_epochs = cloud_epochs[epoch_order]
@@ -165,7 +181,7 @@ def load_snap_clouds(path, dataset, only_last=False):
             Confs[c_i] += np.loadtxt(conf_file, dtype=np.int32)
 
         else:
-            for f in listdir(cloud_folder):
+            for f in listdir_str(cloud_folder):
                 if f.endswith('.ply') and not f.endswith('sub.ply'):
                     data = read_ply(join(cloud_folder, f))
                     labels = data['class']
@@ -176,7 +192,7 @@ def load_snap_clouds(path, dataset, only_last=False):
 
         # Erase ply to save disk memory
         if c_i < len(cloud_folders) - 1:
-            for f in listdir(cloud_folder):
+            for f in listdir_str(cloud_folder):
                 if f.endswith('.ply'):
                     remove(join(cloud_folder, f))
 
@@ -221,7 +237,7 @@ 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 for f in listdir_str(path)]) or ('val_confs.txt' in [f for f in listdir_str(path)]):
             config = Config()
             config.load(path)
         else:
@@ -704,7 +720,7 @@ def experiment_name_1():
     res_path = 'results'
 
     # Gather logs and sort by date
-    logs = np.sort([join(res_path, l) for l in listdir(res_path) if start <= l <= end])
+    logs = np.sort([join(res_path, l) for l in listdir_str(res_path) if start <= l <= end])
 
     # Give names to the logs (for plot legends)
     logs_names = ['name_log_1',
@@ -733,7 +749,7 @@ def experiment_name_2():
     res_path = 'results'
 
     # Gather logs and sort by date
-    logs = np.sort([join(res_path, l) for l in listdir(res_path) if start <= l <= end])
+    logs = np.sort([join(res_path, l) for l in listdir_str(res_path) if start <= l <= end])
 
     # Optionally add a specific log at a specific place in the log list
     logs = logs.astype('<U50')

train_S3DIS.py

@@ -66,30 +66,6 @@ class S3DISConfig(Config):
     #########################
 
     # # Define layers
-    # architecture = ['simple',
-    #                 'resnetb',
-    #                 'resnetb_strided',
-    #                 'resnetb',
-    #                 'resnetb',
-    #                 'resnetb_strided',
-    #                 'resnetb_deformable',
-    #                 'resnetb_deformable',
-    #                 'resnetb_deformable_strided',
-    #                 'resnetb_deformable',
-    #                 'resnetb_deformable',
-    #                 'resnetb_deformable_strided',
-    #                 'resnetb_deformable',
-    #                 'resnetb_deformable',
-    #                 'nearest_upsample',
-    #                 'unary',
-    #                 'nearest_upsample',
-    #                 'unary',
-    #                 'nearest_upsample',
-    #                 'unary',
-    #                 'nearest_upsample',
-    #                 'unary']
-
-    # Define layers
     architecture = ['simple',
                     'resnetb',
                     'resnetb_strided',
@@ -99,11 +75,11 @@ class S3DISConfig(Config):
                     'resnetb',
                     'resnetb',
                     'resnetb_strided',
-                    'resnetb',
+                    'resnetb_deformable',
-                    'resnetb',
+                    'resnetb_deformable',
-                    'resnetb_strided',
+                    'resnetb_deformable_strided',
-                    'resnetb',
+                    'resnetb_deformable',
-                    'resnetb',
+                    'resnetb_deformable',
                     'nearest_upsample',
                     'unary',
                     'nearest_upsample',
@@ -113,6 +89,30 @@ class S3DISConfig(Config):
                     'nearest_upsample',
                     'unary']
 
+    # Define layers
+    # architecture = ['simple',
+    #                 'resnetb',
+    #                 'resnetb_strided',
+    #                 'resnetb',
+    #                 'resnetb',
+    #                 'resnetb_strided',
+    #                 'resnetb',
+    #                 'resnetb',
+    #                 'resnetb_strided',
+    #                 'resnetb',
+    #                 'resnetb',
+    #                 'resnetb_strided',
+    #                 'resnetb',
+    #                 'resnetb',
+    #                 'nearest_upsample',
+    #                 'unary',
+    #                 'nearest_upsample',
+    #                 'unary',
+    #                 'nearest_upsample',
+    #                 'unary',
+    #                 'nearest_upsample',
+    #                 'unary']
+
     ###################
     # KPConv parameters
     ###################
@@ -121,7 +121,7 @@ class S3DISConfig(Config):
     num_kernel_points = 15
 
     # Radius of the input sphere (decrease value to reduce memory cost)
-    in_radius = 1.8
+    in_radius = 1.2
 
     # Size of the first subsampling grid in meter (increase value to reduce memory cost)
     first_subsampling_dl = 0.03
@@ -130,7 +130,7 @@ class S3DISConfig(Config):
     conv_radius = 2.5
 
     # Radius of deformable convolution in "number grid cell". Larger so that deformed kernel can spread out
-    deform_radius = 6.0
+    deform_radius = 5.0
 
     # Radius of the area of influence of each kernel point in "number grid cell". (1.0 is the standard value)
     KP_extent = 1.2