diff --git a/SMIL_torch_batch.py b/SMIL_torch_batch.py
index fdeb69c..8d8290c 100644
--- a/SMIL_torch_batch.py
+++ b/SMIL_torch_batch.py
@@ -188,6 +188,7 @@ def forward(self, beta, pose, trans=None, simplify=False):
                     G[self.parent[i]],
                     self.rotate_translate(R_cube[:, i], J[:, i] - J[:, self.parent[i]])))
         G = torch.stack(G, 1)
+        Jtr = G[..., :4, 3].clone()
         G = G - self.pack(torch.matmul(G, torch.cat([J, J.new_zeros(1).expand(*J.shape[:2], 1)], dim=2).unsqueeze(-1)))
 
         # T = torch.tensordot(self.weights, G, dims=([1], [1]))
@@ -197,7 +198,6 @@ def forward(self, beta, pose, trans=None, simplify=False):
         T = torch.tensordot(G, self.weights, dims=([1], [1])).permute(0, 3, 1, 2)
 
         v = torch.matmul(T, torch.reshape(rest_shape_h, (batch_size, -1, 4, 1))).reshape(batch_size, -1, 4)
-        Jtr = self.regress_joints(v)
 
         if trans is not None:
             trans = trans.unsqueeze(1)