merged

dataset/shapenet.py

@@ -17,24 +17,28 @@ def save_names(name_list, path):
     with open(path, 'wb') as f:
         f.writelines(name_list)
 
+
 class CustomShapeNet(InMemoryDataset):
 
     categories = {key: val for val, key in enumerate(['Box', 'Cone', 'Cylinder', 'Sphere'])}
+    modes = {key: val for val, key in enumerate(['train', 'test', 'predict'])}
 
-    def __init__(self, root, collate_per_segment=True, train=True, transform=None, pre_filter=None, pre_transform=None,
+    def __init__(self, root_dir, collate_per_segment=True, mode='train', transform=None, pre_filter=None,
-                 headers=True, **kwargs):
+                 pre_transform=None, headers=True, has_variations=False, refresh=False, labels_within=False):
-        self.has_headers = headers
+        assert mode in self.modes.keys(), f'"mode" must be one of {self.modes.keys()}'
-        self.collate_per_element = collate_per_segment
+        assert not (collate_per_segment and has_variations), 'Either use each element or pointclouds - with variations'
-        self.train = train
+
-        super(CustomShapeNet, self).__init__(root, transform, pre_transform, pre_filter)
+        #Set the Dataset Parameters
-        path = self.processed_paths[0] if train else self.processed_paths[-1]
+        self.has_headers, self.has_variations, self.labels_within = headers, has_variations, labels_within
-        self.data, self.slices = torch.load(path)
+        self.collate_per_element, self.mode, self.refresh = collate_per_segment, mode, refresh
+        super(CustomShapeNet, self).__init__(root_dir, transform, pre_transform, pre_filter)
+        self.data, self.slices = self._load_dataset()
         print("Initialized")
 
     @property
     def raw_file_names(self):
         # Maybe add more data like validation sets
-        return ['train', 'test']
+        return list(self.modes.keys())
 
     @property
     def processed_file_names(self):
@@ -53,9 +57,18 @@ class CustomShapeNet(InMemoryDataset):
 
     def _load_dataset(self):
         data, slices = None, None
+        filepath = self.processed_paths[self.modes[self.mode]]
+        if self.refresh:
+            try:
+                os.remove(filepath)
+                print('Processed Location "Refreshed" (We deleted the Files)')
+            except FileNotFoundError:
+                print('You meant to refresh the allready processed dataset, but there were none...')
+                print('continue processing')
+                pass
+
         while True:
             try:
-                filepath = os.path.join(self.root, self.processed_dir, f'{"train" if self.train else "test"}.pt')
                 data, slices = torch.load(filepath)
                 print('Dataset Loaded')
                 break
@@ -77,7 +90,7 @@ class CustomShapeNet(InMemoryDataset):
 
     def process(self, delimiter=' '):
         datasets = defaultdict(list)
-        idx, data_folder = (0, self.raw_file_names[0]) if self.train else (1, self.raw_file_names[1])
+        idx, data_folder = self.modes[self.mode], self.raw_file_names[self.modes[self.mode]]
         path_to_clouds = os.path.join(self.raw_dir, data_folder)
 
         if '.headers' in os.listdir(path_to_clouds):
@@ -88,6 +101,8 @@ class CustomShapeNet(InMemoryDataset):
             pass
 
         for pointcloud in tqdm(os.scandir(path_to_clouds)):
+            if self.has_variations:
+                cloud_variations = defaultdict(list)
             if not os.path.isdir(pointcloud):
                 continue
             data, paths = None, list()
@@ -95,16 +110,11 @@ class CustomShapeNet(InMemoryDataset):
                 paths.extend(glob.glob(os.path.join(pointcloud.path, f'*.{ext}')))
 
             for element in paths:
-                if all([x not in os.path.split(element)[-1] for x in ['pc.dat', 'pc.xyz']]):
+                # This was build to filter all variations that aregreater then 25
-                    # Assign training data to the data container
+                pattern = re.compile('^((6[0-1]|[1-5][0-9])_\w+?\d+?|pc|\d+?_pc)\.(xyz|dat)$')
-                    # Following the original logic;
+                if pattern.match(os.path.split(element)[-1]):
-                    # y should be the label;
+                    continue
-                    # pos should be the six dimensional vector describing: !its pos not points!!
+                else:
-                    # x,y,z,x_rot,y_rot,z_rot
-
-                    # Get the y - Label
-                    y_raw = next(i for i, v in enumerate(self.categories.keys()) if v.lower() in element.lower())
-                    # y_raw = os.path.splitext(element)[0].split('_')[-2]
                     with open(element,'r') as f:
                         if self.has_headers:
                             headers = f.__next__()
@@ -118,21 +128,36 @@ class CustomShapeNet(InMemoryDataset):
                     points = torch.tensor(src, dtype=None).squeeze()
                     if not len(points.shape) > 1:
                         continue
-                    y_all = [y_raw] * points.shape[0]
+                    # Place Fake Labels to hold the given structure
+                    if self.labels_within:
+                        y_all = points[:, -1]
+                        points = points[:, :-1]
+                    else:
+                        # Get the y - Label
+                        y_raw = next(i for i, v in enumerate(self.categories.keys()) if v.lower() in element.lower())
+                        y_all = ([y_raw] if self.mode != 'predict' else [-1]) * points.shape[0]
+
                     y = torch.as_tensor(y_all, dtype=torch.int)
                     if self.collate_per_element:
-                        data = Data(y=y, pos=points[:, :3], points=points, norm=points[:, 3:])
+                        data = Data(y=y, pos=points[:, :3])  # , points=points, norm=points[:, 3:])
                     else:
                         if not data:
                             data = defaultdict(list)
-                        for key, val in dict(y=y, pos=points[:, :3], points=points, norm=points[:, 3:]).items():
+                        # points=points, norm=points[:, 3:]
+                        for key, val in dict(y=y, pos=points[:, :3]).items():
                             data[key].append(val)
 
                     data = self._transform_and_filter(data)
                     if self.collate_per_element:
                         datasets[data_folder].append(data)
+                    if self.has_variations:
+                        cloud_variations[int(os.path.split(element)[-1].split('_')[0])].append(data)
             if not self.collate_per_element:
-                datasets[data_folder].append(Data(**{key: torch.cat(data[key]) for key in data.keys()}))
+                if self.has_variations:
+                    for variation in cloud_variations.keys():
+                        datasets[data_folder].append(Data(**{key: torch.cat(data[key]) for key in data.keys()}))
+                else:
+                    datasets[data_folder].append(Data(**{key: torch.cat(data[key]) for key in data.keys()}))
 
         if datasets[data_folder]:
             os.makedirs(self.processed_dir, exist_ok=True)
@@ -147,15 +172,15 @@ class ShapeNetPartSegDataset(Dataset):
     Resample raw point cloud to fixed number of points.
     Map raw label from range [1, N] to [0, N-1].
     """
-    def __init__(self, root_dir, collate_per_segment=True, train=True, transform=None, npoints=1024, headers=True):
+    def __init__(self, root_dir, npoints=1024, **kwargs):
         super(ShapeNetPartSegDataset, self).__init__()
+        kwargs.update(dict(root_dir=root_dir))
         self.npoints = npoints
-        self.dataset = CustomShapeNet(root=root_dir, collate_per_segment=collate_per_segment,
+        self.dataset = CustomShapeNet(**kwargs)
-                                      train=train, transform=transform, headers=headers)
 
     def __getitem__(self, index):
         data = self.dataset[index]
-        points, labels, _, norm = data.pos, data.y, data.points, data.norm
+        points, labels = data.pos, data.y  # , data.points, data.norm
 
         # Resample to fixed number of points
         try:
@@ -163,14 +188,13 @@ class ShapeNetPartSegDataset(Dataset):
         except ValueError:
             choice = []
 
-        points, labels, norm = points[choice, :], labels[choice], norm[choice]
+        points, labels = points[choice, :], labels[choice]
 
         labels -= 1 if self.num_classes() in labels else 0   # Map label from [1, C] to [0, C-1]
 
         sample = {
             'points': points,  # torch.Tensor (n, 3)
-            'labels': labels,  # torch.Tensor (n,)
+            'labels': labels  # torch.Tensor (n,)
-            'normals': norm  # torch.Tensor (n,)
         }
 
         return sample
@@ -180,144 +204,3 @@ class ShapeNetPartSegDataset(Dataset):
 
     def num_classes(self):
         return self.dataset.num_classes
-
-
-class PredictionShapeNet(InMemoryDataset):
-
-    def __init__(self, root, transform=None, pre_filter=None, pre_transform=None, headers=True, refresh=False):
-        self.has_headers = headers
-        self.refresh = refresh
-        super(PredictionShapeNet, self).__init__(root, transform, pre_transform, pre_filter)
-        path = self.processed_paths[0]
-        self.data, self.slices = self._load_dataset()
-        print("Initialized")
-
-    @property
-    def raw_file_names(self):
-        # Maybe add more data like validation sets
-        return ['predict']
-
-    @property
-    def processed_file_names(self):
-        return [f'{x}.pt' for x in self.raw_file_names]
-
-    def download(self):
-        dir_count = len([name for name in os.listdir(self.raw_dir) if os.path.isdir(os.path.join(self.raw_dir, name))])
-        print(f'{dir_count} folders have been found....')
-        if dir_count:
-            return dir_count
-        raise IOError("No raw pointclouds have been found.")
-
-    @property
-    def num_classes(self):
-        return len(self.categories)
-
-    def _load_dataset(self):
-        data, slices = None, None
-        filepath = os.path.join(self.processed_dir, self.processed_file_names[0])
-        if self.refresh:
-            try:
-                os.remove(filepath)
-                print('Processed Location "Refreshed" (We deleted the Files)')
-            except FileNotFoundError:
-                print('You meant to refresh the allready processed dataset, but there were none...')
-                print('continue processing')
-                pass
-
-        while True:
-            try:
-                data, slices = torch.load(filepath)
-                print('Dataset Loaded')
-                break
-            except FileNotFoundError:
-                self.process()
-                continue
-        return data, slices
-
-    def process(self, delimiter=' '):
-
-        datasets, filenames = defaultdict(list), []
-        path_to_clouds = os.path.join(self.raw_dir, self.raw_file_names[0])
-
-        if '.headers' in os.listdir(path_to_clouds):
-            self.has_headers = True
-        elif 'no.headers' in os.listdir(path_to_clouds):
-            self.has_headers = False
-        else:
-            pass
-
-        for pointcloud in tqdm(os.scandir(path_to_clouds)):
-            if not os.path.isdir(pointcloud):
-                continue
-            full_cloud_pattern = '(^\d+?_|^)pc\.(xyz|dat)'
-            pattern = re.compile(full_cloud_pattern)
-            for file in os.scandir(pointcloud.path):
-                if not pattern.match(file.name):
-                    continue
-                with open(file, 'r') as f:
-                        if self.has_headers:
-                            headers = f.__next__()
-                            # Check if there are no useable nodes in this file, header says 0.
-                            if not int(headers.rstrip().split(delimiter)[0]):
-                                continue
-
-                        # Iterate over all rows
-                        src = [[float(x) if x not in ['-nan(ind)', 'nan(ind)'] else 0
-                                for x in line.rstrip().split(delimiter)[None:None]] for line in f if line != '']
-                points = torch.tensor(src, dtype=None).squeeze()
-                if not len(points.shape) > 1:
-                    continue
-                # pos = points[:, :3]
-                # norm = points[:, 3:]
-                y_fake_all = [-1] * points.shape[0]
-                y = torch.as_tensor(y_fake_all, dtype=torch.int)
-                # points = torch.as_tensor(points, dtype=torch.float)
-                # norm = torch.as_tensor(norm, dtype=torch.float)
-                data = Data(y=y, pos=points[:, :3], points=points, norm=points[:, 3:])
-                # , points=points, norm=points[:3], )
-                # ToDo: ANy filter to apply? Then do it here.
-                if self.pre_filter is not None and not self.pre_filter(data):
-                    data = self.pre_filter(data)
-                    raise NotImplementedError
-                # ToDo: ANy transformation to apply? Then do it here.
-                if self.pre_transform is not None:
-                    data = self.pre_transform(data)
-                    raise NotImplementedError
-                datasets[self.raw_file_names[0]].append(data)
-                filenames.append(file)
-
-            os.makedirs(self.processed_dir, exist_ok=True)
-            torch.save(self.collate(datasets[self.raw_file_names[0]]), self.processed_paths[0])
-            # save_names(filenames)
-
-    def __repr__(self):
-        return f'{self.__class__.__name__}({len(self)})'
-
-
-class PredictNetPartSegDataset(Dataset):
-    """
-    Resample raw point cloud to fixed number of points.
-    Map raw label from range [1, N] to [0, N-1].
-    """
-    def __init__(self, root_dir, num_classes, transform=None, npoints=2048, headers=True, refresh=False):
-        super(PredictNetPartSegDataset, self).__init__()
-        self.npoints = npoints
-        self._num_classes = num_classes
-        self.dataset = PredictionShapeNet(root=root_dir, transform=transform, headers=headers, refresh=refresh)
-
-    def __getitem__(self, index):
-        data = self.dataset[index]
-        points, labels, _, norm = data.pos, data.y, data.points, data.norm
-
-        sample = {
-            'points': points,  # torch.Tensor (n, 3)
-            'labels': labels,  # torch.Tensor (n,)
-            'normals': norm  # torch.Tensor (n,)
-        }
-        return sample
-
-    def __len__(self):
-        return len(self.dataset)
-
-    def num_classes(self):
-        return self._num_classes

main.py

@@ -33,11 +33,15 @@ parser.add_argument('--npoints', type=int, default=1024, help='resample points n
 parser.add_argument('--model', type=str, default='', help='model path')
 parser.add_argument('--nepoch', type=int, default=250, help='number of epochs to train for')
 parser.add_argument('--outf', type=str, default='checkpoint', help='output folder')
+parser.add_argument('--labels_within', type=strtobool, default=False, help='defines the label location')
 parser.add_argument('--batch_size', type=int, default=8, help='input batch size')
 parser.add_argument('--test_per_batches', type=int, default=1000, help='run a test batch per training batches number')
 parser.add_argument('--num_workers', type=int, default=4, help='number of data loading workers')
 parser.add_argument('--headers', type=strtobool, default=True, help='if raw files come with headers')
 parser.add_argument('--collate_per_segment', type=strtobool, default=True, help='whether to look at pointclouds or sub')
+parser.add_argument('--has_variations', type=strtobool, default=False,
+                    help='whether a single pointcloud has variations '
+                         'named int(id)_pc.(xyz|dat) look at pointclouds or sub')
 
 
 opt = parser.parse_args()
@@ -53,28 +57,34 @@ torch.manual_seed(opt.manual_seed)
 torch.cuda.manual_seed(opt.manual_seed)
 
 if __name__ == '__main__':
-
     # Dataset and transform
     print('Construct dataset ..')
-    if True:
-        rot_max_angle = 15
-        trans_max_distance = 0.01
 
-        RotTransform = GT.Compose([GT.RandomRotate(rot_max_angle, 0),
+    rot_max_angle = 15
-                                   GT.RandomRotate(rot_max_angle, 1),
+    trans_max_distance = 0.01
-                                   GT.RandomRotate(rot_max_angle, 2)]
-                                  )
-        TransTransform = GT.RandomTranslate(trans_max_distance)
 
-        train_transform = GT.Compose([GT.NormalizeScale(), RotTransform, TransTransform])
+    RotTransform = GT.Compose([GT.RandomRotate(rot_max_angle, 0),
-        test_transform = GT.Compose([GT.NormalizeScale(), ])
+                               GT.RandomRotate(rot_max_angle, 1),
+                               GT.RandomRotate(rot_max_angle, 2)]
+                              )
 
-    dataset = ShapeNetPartSegDataset(root_dir=opt.dataset, collate_per_segment=opt.collate_per_segment,
+    TransTransform = GT.RandomTranslate(trans_max_distance)
-                                     train=True, transform=train_transform, npoints=opt.npoints, headers=opt.headers)
+    train_transform = GT.Compose([GT.NormalizeScale(), RotTransform, TransTransform])
+    test_transform = GT.Compose([GT.NormalizeScale(), ])
+
+    params = dict(root_dir=opt.dataset,
+                  collate_per_segment=opt.collate_per_segment,
+                  transform=train_transform,
+                  npoints=opt.npoints,
+                  labels_within=opt.labels_within,
+                  has_variations=opt.has_variations,
+                  headers=opt.headers
+                  )
+
+    dataset = ShapeNetPartSegDataset(mode='train', **params)
     dataLoader = DataLoader(dataset, batch_size=opt.batch_size, shuffle=True, num_workers=opt.num_workers)
 
-    test_dataset = ShapeNetPartSegDataset(root_dir=opt.dataset, collate_per_segment=opt.collate_per_segment,
+    test_dataset = ShapeNetPartSegDataset(mode='test', **params)
-                                          train=False, transform=test_transform, npoints=opt.npoints, headers=opt.headers)
     test_dataLoader = DataLoader(test_dataset, batch_size=opt.batch_size, shuffle=True, num_workers=opt.num_workers)
 
     num_classes = dataset.num_classes()
@@ -118,7 +128,8 @@ if __name__ == '__main__':
         print('Epoch {}, total epoches {}'.format(epoch+1, opt.nepoch))
 
         net.train()
-
+        # ToDo: We need different dataloader here to train the network in multiple iterations, maybe move the loop down
+        # for dataloader in ...
         for batch_idx, sample in enumerate(dataLoader):
             # points: (batch_size, n, 3)
             # labels: (batch_size, n)

predict/predict.py

@@ -2,7 +2,10 @@ import sys
 import os
 import shutil
 import math
-from dataset.shapenet import PredictNetPartSegDataset
+
+sys.path.append(os.path.dirname(os.path.abspath(__file__)) + '/../')  # add project root directory
+
+from dataset.shapenet import ShapeNetPartSegDataset
 from model.pointnet2_part_seg import PointNet2PartSegmentNet
 import torch_geometric.transforms as GT
 import torch
@@ -237,7 +240,7 @@ sys.path.append(os.path.dirname(os.path.abspath(__file__)) + '/../')  # add proj
 parser = argparse.ArgumentParser()
 parser.add_argument('--dataset', type=str, default='data', help='dataset path')
 parser.add_argument('--npoints', type=int, default=2048, help='resample points number')
-parser.add_argument('--model', type=str, default='./checkpoint/seg_model_custom_241.pth', help='model path')
+parser.add_argument('--model', type=str, default='./checkpoint/seg_model_custom_246.pth', help='model path')
 parser.add_argument('--sample_idx', type=int, default=0, help='select a sample to segment and view result')
 opt = parser.parse_args()
 print(opt)
@@ -279,12 +282,12 @@ if __name__ == '__main__':
     print('load dataset ..')
     test_transform = GT.Compose([GT.NormalizeScale(), ])
 
-    test_dataset = PredictNetPartSegDataset(
+    test_dataset = ShapeNetPartSegDataset(
+        mode='predict',
         root_dir=opt.dataset,
-        num_classes=4,
         transform=None,
         npoints=opt.npoints,
-        refresh=True
+        refresh=False
     )
     num_classes = test_dataset.num_classes()
 

vis/show_seg_res.py

@@ -28,10 +28,10 @@ if __name__ == '__main__':
     print('Construct dataset ..')
     test_transform = GT.Compose([GT.NormalizeScale(),])
 
-    test_dataset = PredictNetPartSegDataset(
+    test_dataset = ShapeNetPartSegDataset(
         root_dir=opt.dataset,
         collate_per_segment=False,
-        train=False,
+        mode='predict',
         transform=test_transform,
         npoints=opt.npoints
     )