Update difpoint/inference.py

difpoint/inference.py

@@ -393,8 +393,6 @@ class Inferencer(object):
 
             for key in x_d_i_info:
                 x_d_i_info[key] = torch.tensor(x_d_i_info[key]).cuda()
-            for key in x_s_info:
-                x_s_info[key] = torch.tensor(x_s_info[key]).cuda()
 
             R_d_i = x_d_i_info['R_d']
 
@@ -404,7 +402,7 @@ class Inferencer(object):
             
 
             if self.inf_cfg.flag_relative_motion:
-                R_new = (R_d_i.cpu().numpy() @ R_d_0.permute(0, 2, 1).cpu().numpy()) @ R_s
+                R_new = (R_d_i @ R_d_0.permute(0, 2, 1)) @ R_s
                 delta_new = x_s_info['exp'].reshape(1, 21, -1) + (x_d_i_info['exp'] - x_d_0_info['exp'])
                 scale_new = x_s_info['scale'] * (x_d_i_info['scale'] / x_d_0_info['scale'])
                 t_new = x_s_info['t'] + (x_d_i_info['t'] - x_d_0_info['t'])
@@ -413,15 +411,10 @@ class Inferencer(object):
                 delta_new = x_d_i_info['exp']
                 scale_new = x_s_info['scale']
                 t_new = x_d_i_info['t']
-
             t_new[..., 2] = 0  # zero tz
-            x_c_s = torch.tensor(x_c_s, dtype=torch.float32).cuda()
-            R_new = torch.tensor(R_new, dtype=torch.float32).cuda()
-            delta_new = torch.tensor(delta_new, dtype=torch.float32).cuda()
-            t_new = torch.tensor(t_new, dtype=torch.float32).cuda()
-            scale_new = torch.tensor(scale_new, dtype=torch.float32).cuda()
+            
             x_d_i_new = scale_new * (x_c_s @ R_new + delta_new) + t_new
-            x_d_i_new = x_d_i_new.cpu().numpy()
+    
 
             # Algorithm 1:
             if not self.inf_cfg.flag_stitching and not self.inf_cfg.flag_eye_retargeting and not self.inf_cfg.flag_lip_retargeting: