disable logging metrics debug cluster

temos/model/teach.py

@@ -500,35 +500,36 @@ class TEACH(BaseModel):
         #     # Compute the metrics
         #     # breakpoint()
         #     # output_features_1_T_lst[0]  -> F, feats -> F, xyz
-        if split == "val":# or batch_idx == 0:
-            self.transforms.rots2rfeats.to(latent_vector_0_T.device)
-            self.transforms.rots2joints.to(latent_vector_0_T.device)
+        # if split == "val":# or batch_idx == 0:
+        #     self.transforms.rots2rfeats.to(latent_vector_0_T.device)
+        #     self.transforms.rots2joints.to(latent_vector_0_T.device)
+
+        #     if not self.hparams.losses.loss_on_transition:
+        #         joints_0 = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(x.detach())) for x in output_features_0_T_lst]
+        #         joints_1 = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(x.detach())) for x in output_features_1_T_lst]
+        #         ref_joints_0 = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(x.detach())) for x in input_motion_feats_0_lst]
+        #         ref_joints_1 = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(x.detach())) for x in input_motion_feats_1_lst]
+
+        #         self.metrics_0.update(joints_0.detach(), ref_joints_0.detach(), length_0.detach())
+        #         self.metrics_1.update(joints_1.detach(), ref_joints_1.detach(), length_1.detach())
+
+        #     else:
+        #         # if self.motiondecoder.hparams.prev_data_mode == "hist_frame_outpast":
+        #         #     # remove the hframes
+        #         #     # M0 / M1[hframes:]
+        #         #     breakpoint()
+        #         #     joints_output = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(torch.cat((x, y[hframes:]))))
+        #         #                      for x, y in zip(output_features_0_T_lst, output_features_1_T_with_transition_lst)]
+        #         # else:
+        #         # Evalute on all the joints (concatenation)
+        #         joints_output = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(torch.cat((x.detach(), y.detach()))))
+        #                             for x, y in zip(output_features_0_T_lst, output_features_1_T_with_transition_lst)]
+
+        #         joints_input = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(torch.cat((x.detach(), y.detach()))))
+        #                         for x, y in zip(input_motion_feats_0_lst, input_motion_feats_1_with_transition_lst)]
+
+        #         self.metrics.update(joints_output, joints_input, total_length)
 
-            if not self.hparams.losses.loss_on_transition:
-                joints_0 = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(x.detach())) for x in output_features_0_T_lst]
-                joints_1 = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(x.detach())) for x in output_features_1_T_lst]
-                ref_joints_0 = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(x.detach())) for x in input_motion_feats_0_lst]
-                ref_joints_1 = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(x.detach())) for x in input_motion_feats_1_lst]
-
-                self.metrics_0.update(joints_0.detach(), ref_joints_0.detach(), length_0.detach())
-                self.metrics_1.update(joints_1.detach(), ref_joints_1.detach(), length_1.detach())
-
-            else:
-                # if self.motiondecoder.hparams.prev_data_mode == "hist_frame_outpast":
-                #     # remove the hframes
-                #     # M0 / M1[hframes:]
-                #     breakpoint()
-                #     joints_output = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(torch.cat((x, y[hframes:]))))
-                #                      for x, y in zip(output_features_0_T_lst, output_features_1_T_with_transition_lst)]
-                # else:
-                # Evalute on all the joints (concatenation)
-                joints_output = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(torch.cat((x.detach(), y.detach()))))
-                                    for x, y in zip(output_features_0_T_lst, output_features_1_T_with_transition_lst)]
-
-                joints_input = [self.transforms.rots2joints(self.transforms.rots2rfeats.inverse(torch.cat((x.detach(), y.detach()))))
-                                for x, y in zip(input_motion_feats_0_lst, input_motion_feats_1_with_transition_lst)]
-
-                self.metrics.update(joints_output, joints_input, total_length)
         # render_list_train = ['1574-1', '7286-0', '6001-0', '4224-2', '3415-0', '2634-0', '2424-1', '4550-0']
         # render_list_val = ['2307-1', '6078-0', '5210-0', '12255-0', '11346-2', '11671-1', '443-8', '3290-3', '2014-0', '973-12']
         # breakpoint()