classdom support get_hierarchy_info

dsdl/dataset/wrapper_dataset.py

@@ -8,9 +8,23 @@
 from .base_dataset import Dataset
 from yaml import load as yaml_load
 from typing import Sequence, Union, Iterable
-from torch.utils.data import Dataset as _Dataset
-from torch.utils.data import DataLoader, ConcatDataset
 from terminaltables import AsciiTable
+try:
+    from torch.utils.data import Dataset as _Dataset
+    from torch.utils.data import DataLoader, ConcatDataset
+except:
+    from ..warning import ImportWarning
+    ImportWarning("'torch' is not installed.")
+
+    class _Dataset:
+        def __init__(self, *args, **kwargs):
+            pass
+    class DataLoader:
+        def __init__(self, *args, **kwargs):
+            pass
+    class ConcatDataset:
+        def __init__(self, *args, **kwargs):
+            pass
 
 try:
     from yaml import CSafeLoader as YAMLSafeLoader

dsdl/geometry/classdomain.py

@@ -1,3 +1,5 @@
+import numpy as np
+
 from dsdl.geometry.registry import CLASSDOMAIN, LABEL
 from dsdl.geometry.label import Label
 from dsdl.geometry.class_domain_attributes import Skeleton
@@ -105,6 +107,38 @@ def get_label(cls, name):
     def get_attribute(cls, attr_name):
         attr_dic = getattr(cls, "__attributes__")
         return attr_dic.get(attr_name, None)
+
+    def get_hierarchy_info(cls):
+        class_names_in = [i.category_name for i in cls.__list__]
+        single_name_dict = {}
+        name_2_index_dict = {}
+        for curr_name in class_names_in:
+            arrs = curr_name.split(".")[1:]
+            for cname in arrs:
+                if cname not in single_name_dict:
+                    single_name_dict[cname] = 0
+        single_class_nums = len(single_name_dict)
+        class_names_sort = sorted(list(single_name_dict.keys()))
+        relation_metric = np.eye(single_class_nums, single_class_nums)
+        for index, key in enumerate(class_names_sort):
+            name_2_index_dict[key] = index
+
+        for idx, item in enumerate(class_names_sort):
+            curr_dict = {}
+            index_list = []
+            for class_name in class_names_in:
+                name_arrs = class_name.split(".")[1:]
+                if item in name_arrs:
+                    curr_index = name_arrs.index(item)
+                    curr_used_names = name_arrs[:curr_index+1]
+                    for used_name in curr_used_names:
+                        if used_name not in curr_dict:
+                            curr_dict[used_name] = 0
+            for ckey in curr_dict:
+                m_index = name_2_index_dict[ckey]
+                index_list.append(m_index)
+                relation_metric[idx, m_index] = 1
+        return class_names_sort, relation_metric
 
 
 def ClassDomain(name, **kwargs):

dsdl/geometry/label.py

@@ -90,6 +90,10 @@ def category_name(self):
         """
         return self._name
 
+    @property
+    def leaf_node_name(self):
+        return self._name.split(".")[-1]
+
     @property
     def openmmlabformat(self):
         """

dsdl/parser/parser.py

@@ -291,8 +291,28 @@ def _parse(self, data_file: str, library_path: str):
                 FIELD_PARSER = ParserField(self.struct_name_params, self.struct_name,struct_params)
                 field_list = dict()
                 for raw_field in define_value["$fields"].items():
-                    field_name = raw_field[0].strip()
-                    field_type = raw_field[1].strip()
+                    try:
+                        field_name = raw_field[0].strip()
+                    except:
+                        msg = f"field name must be string, got `{raw_field[0]}` (type: {type(raw_field[0])})"
+                        if self.report_flag:
+                            temp_check_item = CheckLogItem(
+                                def_name=TypeEnum.STRUCT.value, msg=msg
+                            )
+                            CHECK_LOG.sub_struct.append(temp_check_item)
+                            return
+                        raise Exception(msg)
+                    try:
+                        field_type = raw_field[1].strip()
+                    except:
+                        msg = f"field type must be string, got `{raw_field[1]}` (type: {type(raw_field[1])})"
+                        if self.report_flag:
+                            temp_check_item = CheckLogItem(
+                                def_name=TypeEnum.STRUCT.value, msg=msg
+                            )
+                            CHECK_LOG.sub_struct.append(temp_check_item)
+                            return
+                        raise Exception(msg)
                     # 判断field_name是否为python保留字和是符合命名规范
                     try:
                         check_name_format(field_name)

setup.py

@@ -44,6 +44,7 @@ def readme():
         "tqdm>=4.65.0",
         "scikit-image>=0.19.3",
         "tifffile"
+        "terminaltables>=3.1.10"
     ],
     classifiers=[
         "Programming Language :: Python :: 3.8",