Merge pull request #18 from uug-ai/fix_bug

Fix occasional crash bug due to different video resolutions

Author: cedricve

projects/base_project.py

@@ -6,7 +6,11 @@
 )
 from projects.ibase_project import IBaseProject
 from utils.VariableClass import VariableClass
+from ultralytics import YOLO
+
 import yaml
+import os
+import torch
 
 
 class BaseProject(IBaseProject):
@@ -24,6 +28,7 @@ def __init__(self):
         self.proj_dir = None
         self.mapping = None
         self.device = None
+        self.models = []
 
     def condition_func(self, total_results):
         """
@@ -59,6 +64,9 @@ def connect_models(self):
         raise NotImplemented('Should override this!!!')
 
     def __read_config__(self, path):
+        """
+        See ibase_project.py
+        """
         with open(path, 'r') as file:
             config = yaml.safe_load(file)
 
@@ -72,3 +80,27 @@ def __read_config__(self, path):
 
         raise TypeError('Error while reading configuration file, '
                         'make sure models and allowed_classes have the same size')
+
+    def __connect_models__(self):
+        """
+        See ibase_project.py
+        """
+        _cur_dir = os.getcwd()
+        # initialise the yolo model, additionally use the device parameter to specify the device to run the model on.
+        self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
+        _cur_dir = pdirname(pabspath(__file__))
+        model_dir = pjoin(_cur_dir, f'../models')
+        model_dir = pabspath(model_dir)  # normalise the link
+
+        models = []
+        for model_name in self._config.get('models'):
+            model = YOLO(pjoin(model_dir, model_name)).to(self.device)
+            models.append(model)
+
+        return models
+
+    def reset_models(self):
+        """
+        See ibase_project.py
+        """
+        self.models = self.__connect_models__()

projects/helmet/helmet_project.py

@@ -1,17 +1,6 @@
-from os.path import (
-    join as pjoin,
-    dirname as pdirname,
-    abspath as pabspath
-)
-
-from ultralytics import YOLO
-
 from projects.base_project import BaseProject
 from projects.helmet.ihelmet_project import IHelmetProject
 
-import os
-import torch
-
 config_path = './projects/helmet/helmet_config.yaml'
 
 
@@ -117,24 +106,14 @@ def connect_models(self):
         Initializes the YOLO models and connects them to the appropriate device (CPU or GPU).
 
         Returns:
-            tuple: A tuple containing two YOLO models.
+            models: A tuple containing two YOLO models.
+            models_allowed_classes: List of corresponding allowed classes for each model.
 
         Raises:
             ModuleNotFoundError: If the models cannot be loaded.
         """
 
-        _cur_dir = os.getcwd()
-        # initialise the yolo model, additionally use the device parameter to specify the device to run the model on.
-        self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
-        _cur_dir = pdirname(pabspath(__file__))
-        model_dir = pjoin(_cur_dir, f'../../models')
-        model_dir = pabspath(model_dir)  # normalise the link
-
-        models = []
-        for model_name in self._config.get('models'):
-            model = YOLO(pjoin(model_dir, model_name)).to(self.device)
-            models.append(model)
-
+        models = self.__connect_models__()
         models_allowed_classes = self._config.get('allowed_classes')
 
         if not models:

projects/ibase_project.py

@@ -40,3 +40,37 @@ def create_proj_save_dir(self):
         Create project save directory after initializing the project.
         """
         pass
+
+    @abstractmethod
+    def __read_config__(self, path):
+        """
+        Read project's configuration file.
+
+        Returns:
+            tuple: Configuration file in dictionary format.
+
+        Raises:
+            TypeError: If the models cannot be loaded.
+        """
+        pass
+
+    @abstractmethod
+    def __connect_models__(self):
+        """
+        Initializes the YOLO models and connects them to the appropriate device (CPU or GPU).
+
+        Returns:
+            tuple: A tuple containing two YOLO models.
+
+        Raises:
+            ModuleNotFoundError: If the models cannot be loaded.
+        """
+        pass
+
+    @abstractmethod
+    def reset_models(self):
+        """
+        Reset model after processing video to avoid memory allocation error when the upcoming video comes in with
+        different resolution.
+        """
+        pass

projects/person/person_project.py

@@ -1,17 +1,6 @@
-from os.path import (
-    join as pjoin,
-    dirname as pdirname,
-    abspath as pabspath
-)
-
-from ultralytics import YOLO
-
 from projects.base_project import BaseProject
 from projects.person.iperson_project import IPersonProject
 
-import os
-import torch
-
 config_path = './projects/person/person_config.yaml'
 
 
@@ -111,30 +100,19 @@ def connect_models(self):
         Initializes the YOLO models and connects them to the appropriate device (CPU or GPU).
 
         Returns:
-            tuple: A tuple containing two YOLO models.
+            models: A tuple containing two YOLO models.
+            models_allowed_classes: List of corresponding allowed classes for each model.
 
         Raises:
             ModuleNotFoundError: If the models cannot be loaded.
         """
 
-        _cur_dir = os.getcwd()
-        # initialise the yolo model, additionally use the device parameter to specify the device to run the model on.
-        self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
-        _cur_dir = pdirname(pabspath(__file__))
-        model_dir = pjoin(_cur_dir, f'../../models')
-        model_dir = pabspath(model_dir)  # normalise the link
-
-        models = []
-        for model_name in self._config.get('models'):
-            model = YOLO(pjoin(model_dir, model_name)).to(self.device)
-            models.append(model)
-
+        models = self.__connect_models__()
         models_allowed_classes = self._config.get('allowed_classes')
 
         if not models:
             raise ModuleNotFoundError('Model not found!')
 
         print(f'1. Using device: {self.device}')
-        print(
-            f"2. Using {len(models)} models: {[model_name for model_name in self._config.get('models')]}")
+        print(f"2. Using {len(models)} models: {[model_name for model_name in self._config.get('models')]}")
         return models, models_allowed_classes

services/harvest_service.py

@@ -190,6 +190,7 @@ def evaluate(self, video):
                     frame,
                     skip_frames_counter)
             # Free all resources
+            self.project.reset_models()
             cv2.destroyAllWindows()
 
         return self.export.result_dir_path