diff --git a/mmpose/models/heads/regression_heads/temporal_regression_head.py b/mmpose/models/heads/regression_heads/temporal_regression_head.py
index 61e585103f..902be8099e 100644
--- a/mmpose/models/heads/regression_heads/temporal_regression_head.py
+++ b/mmpose/models/heads/regression_heads/temporal_regression_head.py
@@ -1,11 +1,10 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from typing import Optional, Sequence, Tuple, Union
 
-import numpy as np
 import torch
 from torch import Tensor, nn
 
-from mmpose.evaluation.functional import keypoint_pck_accuracy
+from mmpose.evaluation.functional import keypoint_mpjpe
 from mmpose.registry import KEYPOINT_CODECS, MODELS
 from mmpose.utils.tensor_utils import to_numpy
 from mmpose.utils.typing import (ConfigType, OptConfigType, OptSampleList,
@@ -133,14 +132,13 @@ def loss(self,
         losses.update(loss_pose3d=loss)
 
         # calculate accuracy
-        _, avg_acc, _ = keypoint_pck_accuracy(
+        mpjpe_err = keypoint_mpjpe(
             pred=to_numpy(pred_outputs),
             gt=to_numpy(lifting_target_label),
-            mask=to_numpy(lifting_target_weight) > 0,
-            thr=0.05,
-            norm_factor=np.ones((pred_outputs.size(0), 3), dtype=np.float32))
+            mask=to_numpy(lifting_target_weight) > 0)
 
-        mpjpe_pose = torch.tensor(avg_acc, device=lifting_target_label.device)
+        mpjpe_pose = torch.tensor(
+            mpjpe_err, device=lifting_target_label.device)
         losses.update(mpjpe=mpjpe_pose)
 
         return losses
diff --git a/mmpose/models/heads/regression_heads/trajectory_regression_head.py b/mmpose/models/heads/regression_heads/trajectory_regression_head.py
index a1608aaae7..b4d02f2ce3 100644
--- a/mmpose/models/heads/regression_heads/trajectory_regression_head.py
+++ b/mmpose/models/heads/regression_heads/trajectory_regression_head.py
@@ -1,11 +1,10 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from typing import Optional, Sequence, Tuple, Union
 
-import numpy as np
 import torch
 from torch import Tensor, nn
 
-from mmpose.evaluation.functional import keypoint_pck_accuracy
+from mmpose.evaluation.functional import keypoint_mpjpe
 from mmpose.registry import KEYPOINT_CODECS, MODELS
 from mmpose.utils.tensor_utils import to_numpy
 from mmpose.utils.typing import (ConfigType, OptConfigType, OptSampleList,
@@ -132,14 +131,13 @@ def loss(self,
         losses.update(loss_traj=loss)
 
         # calculate accuracy
-        _, avg_acc, _ = keypoint_pck_accuracy(
+        mpjpe_err = keypoint_mpjpe(
             pred=to_numpy(pred_outputs),
             gt=to_numpy(lifting_target_label),
-            mask=to_numpy(trajectory_weights) > 0,
-            thr=0.05,
-            norm_factor=np.ones((pred_outputs.size(0), 3), dtype=np.float32))
+            mask=to_numpy(trajectory_weights) > 0)
 
-        mpjpe_traj = torch.tensor(avg_acc, device=lifting_target_label.device)
+        mpjpe_traj = torch.tensor(
+            mpjpe_err, device=lifting_target_label.device)
         losses.update(mpjpe_traj=mpjpe_traj)
 
         return losses
diff --git a/mmpose/models/utils/rtmcc_block.py b/mmpose/models/utils/rtmcc_block.py
index bd4929454c..fea4edcb3a 100644
--- a/mmpose/models/utils/rtmcc_block.py
+++ b/mmpose/models/utils/rtmcc_block.py
@@ -198,13 +198,15 @@ def __init__(self,
 
         nn.init.xavier_uniform_(self.uv.weight)
 
-        if act_fn == 'SiLU':
+        if act_fn == 'SiLU' or act_fn == nn.SiLU:
             assert digit_version(TORCH_VERSION) >= digit_version('1.7.0'), \
                 'SiLU activation requires PyTorch version >= 1.7'
 
             self.act_fn = nn.SiLU(True)
-        else:
+        elif act_fn == 'ReLU' or act_fn == nn.ReLU:
             self.act_fn = nn.ReLU(True)
+        else:
+            raise NotImplementedError
 
         if in_token_dims == out_token_dims:
             self.shortcut = True