diff --git a/stable_baselines3/common/buffers.py b/stable_baselines3/common/buffers.py
index 16f206f..09b80c3 100644
--- a/stable_baselines3/common/buffers.py
+++ b/stable_baselines3/common/buffers.py
@@ -121,7 +121,7 @@ class BaseBuffer(ABC):
         """
         raise NotImplementedError()
 
-    def to_torch(self, array: np.ndarray, copy: bool = False) -> th.Tensor:
+    def to_torch(self, array: np.ndarray, copy: bool = True) -> th.Tensor:
         """
         Convert a numpy array to a PyTorch tensor.
         Note: it copies the data by default
@@ -132,8 +132,8 @@ class BaseBuffer(ABC):
         :return:
         """
         if copy:
-            return th.tensor(array).to(self.device, memory_format=th.channels_last, non_blocking=True)
-        return th.as_tensor(array).to(self.device, memory_format=th.channels_last, non_blocking=True)
+            return th.tensor(array).to(self.device, non_blocking=True)
+        return th.as_tensor(array).to(self.device, non_blocking=True)
 
     @staticmethod
     def _normalize_obs(
diff --git a/stable_baselines3/common/off_policy_algorithm.py b/stable_baselines3/common/off_policy_algorithm.py
index 20b838e..5e018fc 100644
--- a/stable_baselines3/common/off_policy_algorithm.py
+++ b/stable_baselines3/common/off_policy_algorithm.py
@@ -218,7 +218,7 @@ class OffPolicyAlgorithm(BaseAlgorithm):
             self.lr_schedule,
             **self.policy_kwargs,  # pytype:disable=not-instantiable
         )
-        self.policy = self.policy.to(self.device, memory_format=th.channels_last, non_blocking=True)
+        self.policy = self.policy.to(self.device)
 
         # Convert train freq parameter to TrainFreq object
         self._convert_train_freq()
diff --git a/stable_baselines3/common/on_policy_algorithm.py b/stable_baselines3/common/on_policy_algorithm.py
index 8ce75f9..35ad2b9 100644
--- a/stable_baselines3/common/on_policy_algorithm.py
+++ b/stable_baselines3/common/on_policy_algorithm.py
@@ -121,7 +121,7 @@ class OnPolicyAlgorithm(BaseAlgorithm):
             use_sde=self.use_sde,
             **self.policy_kwargs  # pytype:disable=not-instantiable
         )
-        self.policy = self.policy.to(self.device, memory_format=th.channels_last, non_blocking=True)
+        self.policy = self.policy.to(self.device)
 
     def collect_rollouts(
         self,
diff --git a/stable_baselines3/common/utils.py b/stable_baselines3/common/utils.py
index b713a28..b7c4cb9 100644
--- a/stable_baselines3/common/utils.py
+++ b/stable_baselines3/common/utils.py
@@ -461,12 +461,9 @@ def obs_as_tensor(
     :return: PyTorch tensor of the observation on a desired device.
     """
     if isinstance(obs, np.ndarray):
-        return th.as_tensor(obs).to(device, memory_format=th.channels_last, non_blocking=True)
+        return th.as_tensor(obs).to(device, non_blocking=True)
     elif isinstance(obs, dict):
-        return {
-            key: th.as_tensor(_obs).to(device, memory_format=th.channels_last, non_blocking=True)
-            for (key, _obs) in obs.items()
-        }
+        return {key: th.as_tensor(_obs).to(device, non_blocking=True) for (key, _obs) in obs.items()}
     else:
         raise Exception(f"Unrecognized type of observation {type(obs)}")