feat(layout): More robust formatting of the results returned by gpt.

README.md

@@ -96,7 +96,7 @@ CUDA_VISIBLE_DEVICES=0 nohup python apps/image_to_3d.py > /dev/null 2>&1 &
 Generate physically plausible 3D assets from image input via the command-line API.
 ```sh
 img3d-cli --image_path apps/assets/example_image/sample_00.jpg apps/assets/example_image/sample_01.jpg \
---n_retry 1 --output_root outputs/imageto3d
+--n_retry 2 --output_root outputs/imageto3d
 
 # See result(.urdf/mesh.obj/mesh.glb/gs.ply) in ${output_root}/sample_xx/result
 ```
@@ -246,7 +246,7 @@ Remove `--insert_robot` if you don't consider the robot pose in layout generatio
 CUDA_VISIBLE_DEVICES=0 nohup layout-cli \
 --task_descs "apps/assets/example_layout/task_list.txt" \
 --bg_list "outputs/bg_scenes/scene_list.txt" \
---n_image_retry 4 --n_asset_retry 3 --n_pipe_retry 2 \
+--n_image_retry 4 --n_asset_retry 3 --n_pipe_retry 3 \
 --output_root "outputs/layouts_gens" --insert_robot > layouts_gens.log &
 ```
 

docs/tutorials/image_to_3d.md

@@ -10,7 +10,7 @@ Support the use of [SAM3D](https://github.com/facebookresearch/sam-3d-objects) o
 ```bash
 img3d-cli --image_path apps/assets/example_image/sample_00.jpg \
 apps/assets/example_image/sample_01.jpg \
---n_retry 1 --output_root outputs/imageto3d
+--n_retry 2 --output_root outputs/imageto3d
 ```
 
 You will get the following results:

docs/tutorials/layout_gen.md

@@ -60,7 +60,7 @@ You can also run multiple tasks via a task list file in the backend.
 CUDA_VISIBLE_DEVICES=0 nohup layout-cli \
   --task_descs "apps/assets/example_layout/task_list.txt" \
   --bg_list "outputs/bg_scenes/scene_list.txt" \
-  --n_image_retry 4 --n_asset_retry 3 --n_pipe_retry 2 \
+  --n_image_retry 4 --n_asset_retry 3 --n_pipe_retry 3 \
   --output_root "outputs/layouts_gens" \
   --insert_robot > layouts_gens.log &
 ```

embodied_gen/scripts/imageto3d.py

@@ -105,7 +105,7 @@ def parse_args():
     parser.add_argument(
         "--n_retry",
         type=int,
-        default=2,
+        default=3,
     )
     parser.add_argument("--disable_decompose_convex", action="store_true")
     parser.add_argument("--texture_size", type=int, default=2048)
@@ -163,7 +163,7 @@ def entrypoint(**kwargs):
                     outputs = image3d_model_infer(PIPELINE, seg_image, seed)
                 except Exception as e:
                     logger.error(
-                        f"[Pipeline Failed] process {image_path}: {e}, skip."
+                        f"[Image3D Failed] process {image_path}: {e}, retry: {try_idx+1}/{args.n_retry}"
                     )
                     continue
 

embodied_gen/validators/urdf_convertor.py

@@ -204,8 +204,13 @@ def parse_response(self, response: str) -> dict[str, any]:
         Returns:
             dict[str, any]: Parsed attributes.
         """
-        lines = response.split("\n")
-        lines = [line.strip() for line in lines if line]
+        raw_lines = response.split("\n")
+        lines = []
+        for line in raw_lines:
+            line = line.strip()
+            if line and not line.startswith("```") and ":" in line:
+                lines.append(line)
+
         category = lines[0].split(": ")[1]
         description = lines[1].split(": ")[1]
         min_height, max_height = map(