pipeline for single cluster

main_pipeline.py

@@ -72,7 +72,7 @@ if __name__ == '__main__':
     test_dataset = ShapeNetPartSegDataset('data', mode=GlobalVar.data_split.predict, collate_per_segment=False,
                                           refresh=True, transform=transforms)
 
-    grid_clusters = cluster_cubes(test_dataset[1], [3, 3, 3], max_points_per_cluster=1024)
+    grid_clusters = cluster_cubes(test_dataset[1], [1, 1, 1], max_points_per_cluster=32768)
 
     ps.init()
 

models/point_net_2.py

@@ -25,7 +25,7 @@ class PointNet2(BaseValMixin,
         # Dataset
         # =============================================================================
         # rot_max_angle = 15
-        trans_max_distance = 0.01
+        trans_max_distance = 0.02
         transforms = Compose(
             [
                 RandomFlip(0, p=0.8),

utils/pointcloud.py

@@ -103,10 +103,6 @@ def cluster_per_column(pc, column):
 
 
 def cluster_cubes(data, cluster_dims, max_points_per_cluster=-1, min_points_per_cluster=-1):
-    if cluster_dims[0] == 1 and cluster_dims[1] == 1 and cluster_dims[2] == 1:
-        print("no need to cluster.")
-        return [farthest_point_sampling(data, max_points_per_cluster)]
-
     if isinstance(data, Data):
         import torch
         candidate_list = list()
@@ -119,6 +115,10 @@ def cluster_cubes(data, cluster_dims, max_points_per_cluster=-1, min_points_per_
 
         data = torch.cat(candidate_list, dim=-1).numpy()
 
+    if cluster_dims[0] == 1 and cluster_dims[1] == 1 and cluster_dims[2] == 1:
+        print("no need to cluster.")
+        return [farthest_point_sampling(data, max_points_per_cluster)]
+
     max = data[:, :3].max(axis=0)
     max += max * 0.01