update

app.py

@@ -218,7 +218,7 @@ with gr.Blocks(
                 height=300,
             )
             model_output_gs = gr.Model3D(
-                label="Gaussian Representation", height=300, interactive=False,
+                label="Gaussian Representation", height=300, interactive=False
             )
             aligned_gs = gr.Textbox(visible=False)
             with gr.Row():

asset3d_gen/models/delight_model.py

@@ -57,8 +57,8 @@ class DelightingModel(object):
         pipeline.set_progress_bar_config(disable=True)
 
         pipeline.to(self.device, torch.float16)
-        # pipeline.enable_model_cpu_offload()
-        # pipeline.enable_xformers_memory_efficient_attention()
+        pipeline.enable_model_cpu_offload()
+        pipeline.enable_xformers_memory_efficient_attention()
         self.pipeline = pipeline
 
     def recenter_image(

asset3d_gen/models/gs_model.py

@@ -254,6 +254,7 @@ class GaussianBase:
         opacities = numpy_data["_opacities"]
         sh0 = numpy_data["_features_dc"]
         shN = numpy_data.get("_features_rest", np.zeros((means.shape[0], 0)))
+        shN = shN.reshape(means.shape[0], -1)
 
         # Create a mask to identify rows with NaN or Inf in any of the numpy_data arrays  # noqa
         if enable_mask:

asset3d_gen/models/sr_model.py

@@ -33,7 +33,7 @@ class ImageStableSR:
             torch_dtype=torch.float16,
         ).to(device)
         self.up_pipeline_x4.set_progress_bar_config(disable=True)
-        # self.up_pipeline_x4.enable_model_cpu_offload()
+        self.up_pipeline_x4.enable_model_cpu_offload()
 
     def __call__(
         self,

asset3d_gen/models/text_model.py

@@ -75,7 +75,7 @@ def build_text2img_ip_pipeline(
     pipe.set_ip_adapter_scale([ref_scale])
 
     pipe = pipe.to(device)
-    # pipe.enable_model_cpu_offload()
+    pipe.enable_model_cpu_offload()
     # pipe.enable_xformers_memory_efficient_attention()
     # pipe.enable_vae_slicing()
 
@@ -106,8 +106,8 @@ def build_text2img_pipeline(
         force_zeros_for_empty_prompt=False,
     )
     pipe = pipe.to(device)
-    # pipe.enable_model_cpu_offload()
-    # pipe.enable_xformers_memory_efficient_attention()
+    pipe.enable_model_cpu_offload()
+    pipe.enable_xformers_memory_efficient_attention()
 
     return pipe
 

asset3d_gen/models/texture_model.py

@@ -86,6 +86,6 @@ def build_texture_gen_pipe(
         pipe.set_ip_adapter_scale([ip_adapt_scale])
 
     pipe = pipe.to(device)
-    # pipe.enable_model_cpu_offload()
+    pipe.enable_model_cpu_offload()
 
     return pipe

asset3d_gen/validators/urdf_convertor.py

@@ -165,7 +165,6 @@ class URDFGenerator(object):
             "description": description.lower(),
             "min_height": round(min_height, 4),
             "max_height": round(max_height, 4),
-            "real_height": round((min_height + max_height) / 2, 4),
             "min_mass": round(min_mass, 4),
             "max_mass": round(max_mass, 4),
             "mu1": round(mu1, 2),
@@ -376,11 +375,10 @@ class URDFGenerator(object):
         response = self.gpt_client.query(text_prompt, image_path)
         if response is None:
             asset_attrs = {
-                "category": "unknown",
-                "description": "unknown",
+                "category": category.lower(),
+                "description": category.lower(),
                 "min_height": 1,
                 "max_height": 1,
-                "real_height": 1,
                 "min_mass": 1,
                 "max_mass": 1,
                 "mu1": 0.8,
@@ -394,6 +392,10 @@ class URDFGenerator(object):
             if key in kwargs:
                 asset_attrs[key] = kwargs[key]
 
+        asset_attrs["real_height"] = round(
+            (asset_attrs["min_height"] + asset_attrs["max_height"]) / 2, 4
+        )
+
         self.estimated_attrs = self.get_estimated_attributes(asset_attrs)
 
         urdf_path = self.generate_urdf(mesh_path, output_root, asset_attrs)

common.py

@@ -637,7 +637,7 @@ def text2image_fn(
     if postprocess:
         for idx in range(len(images)):
             image = images[idx]
-            images[idx] = preprocess_image_fn(image)
+            images[idx] = preprocess_image_fn(image, req)
 
     save_paths = []
     for idx, image in enumerate(images):

requirements.txt

@@ -41,7 +41,3 @@ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.0.post2/flas
 https://huggingface.co/spaces/xinjjj/ImgRoboAssetGen/resolve/main/wheels/diff_gaussian_rasterization-0.0.0-cp310-cp310-linux_x86_64.whl
 https://huggingface.co/spaces/xinjjj/ImgRoboAssetGen/resolve/main/wheels/nvdiffrast-0.3.3-cp310-cp310-linux_x86_64.whl
 https://huggingface.co/spaces/xinjjj/ImgRoboAssetGen/resolve/main/wheels/kaolin-0.16.0-cp310-cp310-linux_x86_64.whl
-
-# https://huggingface.co/spaces/JeffreyXiang/TRELLIS/resolve/main/wheels/diff_gaussian_rasterization-0.0.0-cp310-cp310-linux_x86_64.whl
-# https://huggingface.co/spaces/JeffreyXiang/TRELLIS/resolve/main/wheels/nvdiffrast-0.3.3-cp310-cp310-linux_x86_64.whl
-# kaolin@git+https://github.com/NVIDIAGameWorks/[email protected]