Update models/depth_normal_pipeline_clip.py

models/depth_normal_pipeline_clip.py

@@ -79,7 +79,6 @@ class DepthNormalEstimationPipeline(DiffusionPipeline):
                  match_input_res:bool =True,
                  batch_size:int = 0,
                  domain: str = "indoor",
-                 #seed: int = 0,
                  color_map: str="Spectral",
                  show_progress_bar:bool = True,
                  ensemble_kwargs: Dict = None,
@@ -148,7 +147,6 @@ class DepthNormalEstimationPipeline(DiffusionPipeline):
                 input_rgb=batched_image,
                 num_inference_steps=denoising_steps,
                 domain=domain,
-                #seed=seed,
                 show_pbar=show_progress_bar,
             )
             depth_pred_ls.append(depth_pred_raw.detach().clone())
@@ -232,7 +230,6 @@ class DepthNormalEstimationPipeline(DiffusionPipeline):
     def single_infer(self,input_rgb:torch.Tensor,
                      num_inference_steps:int,
                      domain:str,
-                     #seed: int,
                      show_pbar:bool,):
 
         device = input_rgb.device
@@ -244,9 +241,7 @@ class DepthNormalEstimationPipeline(DiffusionPipeline):
         # encode image
         rgb_latent = self.encode_RGB(input_rgb)
         
-        # Initial depth map (Guassian noise)
-        #if seed >= 0:
-            #torch.manual_seed(0)
+        # Initial geometric maps (Guassian noise)
         geo_latent = torch.randn(rgb_latent.shape, device=device, dtype=self.dtype).repeat(2,1,1,1)
         rgb_latent = rgb_latent.repeat(2,1,1,1)
 
@@ -258,7 +253,7 @@ class DepthNormalEstimationPipeline(DiffusionPipeline):
             (rgb_latent.shape[0], 1, 1)
         )  # [B, 1, 768]
 
-        # hybrid hierarchical switcher 
+        # hybrid switcher 
         geo_class = torch.tensor([[0., 1.], [1, 0]], device=device, dtype=self.dtype)
         geo_embedding = torch.cat([torch.sin(geo_class), torch.cos(geo_class)], dim=-1)