feat: Add MoGe-2 surface generation and export functionality

- Introduced new enum items for depth plane and surface models with caching. - Added MOGE2_MESH_JOB to operators for converting images to surfaces. - Implemented surface mesh building and exporting in RunJob operator. - Enhanced UI to include export format options for MoGe-2. - Created validation tools for MoGe-2 surface assets and depth planes. - Updated geometry nodes to support new surface generation features. - Refactored existing tests to accommodate new functionality and ensure stability.
2026-08-13 22:07:36 +08:00
parent a6370d678d
commit caeb6d706c
17 changed files with 902 additions and 251 deletions
--- a/README.md
+++ b/README.md
@@ -20,7 +20,7 @@ SnapForm 是一个 Blender 扩展，可以把一张照片、一段视频或一
 5. 打开 3D 视图右侧的 `SnapForm` 面板，点击 `Install Environment`

 DA3、MoGe-2、utils3d、BEN2 源码包和 uv 已内置在扩展里，安装时直接从本地解包，不需要访问 GitHub。首次使用仍会自动
-下载运行环境依赖（PyTorch 等），随后自动下载 `DA3 Small` 和 `BEN2 Base`。其他 AI 模型可在
+下载运行环境依赖（PyTorch 等），随后自动下载 `MoGe-2 ViT-B Normal` 和 `BEN2 Base`。其他 AI 模型可在
 `Models` panel 中按需下载，请保持网络畅通，首次安装需要一些时间。

 ## 快速上手
@@ -44,14 +44,16 @@ DA3、MoGe-2、utils3d、BEN2 源码包和 uv 已内置在扩展里，安装时

 ### 转换为透视网格

-选中单张图片的 Image Empty，右键打开 `SnapForm` 子菜单并选择 `Convert to Surface`。SnapForm 会按 DA3
-相机内参与深度逐像素反投影，并把生成网格匹配到原 Empty 的可见尺寸、Offset 与 Transform。DA3
-confidence 与边拉伸结果会保存为 Surface attributes，再由一个可复用的 Geometry Nodes
+选中单张图片的 Image Empty，右键打开 `SnapForm` 子菜单并选择 `Convert to Surface`。可选择 MoGe-2
+ViT-S、ViT-B、ViT-L 或 DA3 Small、Base，其中 MoGe-2 ViT-B 为默认模型。SnapForm 会按模型输出的
+相机内参与三维点建立透视网格，并匹配原 Empty 的可见尺寸、Offset 与 Transform。MoGe-2 直接使用官方
+`utils3d.build_mesh_from_map` 与深度断层 mask 生成干净拓扑；DA3 保留 confidence 与边拉伸剔除。两者再由独立的 Geometry Nodes
 修改器完成剔面、快速重拓扑、Relax 与 UV 重投。开启 Remove Background 时，背景三角面会在转换期间
 直接删除，不再交给 Geometry Nodes。匹配相机默认不创建；开启后会作为 Surface 的子级。
 当前此功能仅支持单张图片。

-转换对话框提供模型与 `Resolution Scale`、`Create Matching Camera`、`Remove Background`、`Unshaded`，
+转换对话框提供模型与 `Create Matching Camera`、`Remove Background`、`Unshaded`；DA3 显示
+`Resolution Scale`，MoGe-2 显示 `Resolution Level`，
 以及分割线下方的 `Keep Original`。
 界面会显示计算所得的实际 Process Res。剔除和重拓扑参数直接在
 Geometry Nodes 修改器中调整；`Unshaded` 可使用与 Depth Plane 相同的 Emission 材质。普通材质的
--- a/docs/technical.md
+++ b/docs/technical.md
@@ -25,7 +25,7 @@
 - Environment 已安装时不额外占用面板空间；模型列表每个模型只占一行
 - `Debug` 默认折叠，内部的 `DA3 Output Test`、`MoGe-2 API Test` 与 `BEN2 Removal Test` 也分别默认折叠
 - 使用一个输出类型下拉菜单和一个 `Generate Output` 按钮
- 只显示当前格式的专属参数；GLB 与 PLY 设置不会同时出现
+- DA3 与 MoGe-2 Debug 都先选择单一 Output Format，只显示当前格式的专属参数；GLB 与 PLY 不会在一次测试中同时生成
 - Debug 的模型选项按格式过滤；PLY 只提供两个带 Gaussian head 的 Giant 模型
 - Debug 与 Convert to Depth Plane 的模型选项直接显示下载状态；未下载模型带
  `(Missing)`，已下载模型保持原名
@@ -110,7 +110,7 @@ Cyclic 沿用原 Empty 的设置。序列帧无法被 Blender 整体 Pack，因
     `Models` panel 下载所选模型，并固定使用 `process_res_method="upper_bound_resize"`；设备和参考视图策略
     继续使用场景设置。Model 下拉菜单只列出已经下载的 DA3 模型；若没有可用模型，对话框会引导到
     `Models` panel 下载。
-   - `MoGe-2`：使用 Base Normal 模型输出 metric depth、normal、validity mask 与 normalized intrinsics。正式转换的 `results.npz` 不保存体积较大的 points；Debug 测试会保留 points，并另外写出 depth、normal、mask 预览。
+   - `MoGe-2`：使用所选模型输出 metric depth、normal、validity mask 与 normalized intrinsics。Depth Plane 不保存体积较大的 points；Surface 会在 AI 进程内用 points 和官方 `utils3d.build_mesh_from_map` 直接生成 `mesh.npz`。Debug 始终保留原始 `results.npz`，并按 Output Format 单独生成 Maps、纹理 GLB 或彩色 PLY。
   - `Original`：直接把原图作为深度贴图，不显示估算参数。仅在开启 `Remove Background` 时运行 BEN2
      取得前景 Alpha；视频或序列帧按需同时抽取原始帧与前景序列，把原始帧作为 `.depth` 序列。
   `Original` 是默认深度来源。选择 `DA3` 时，其 Model、Use Ray Pose 与 Resolution Scale
@@ -168,8 +168,8 @@ Cyclic 沿用原 Empty 的设置。序列帧无法被 Blender 整体 Pack，因
 1. 在 3D View 选中仅包含一帧的 Image Empty，右键打开 `SnapForm` 子菜单并选择
   `Convert to Surface`。该入口独立于
   `Convert to Depth Plane`，当前不处理视频或序列帧。
-   对话框中的 Resolution Scale 与 Depth Plane 使用相同设计：原图长边乘以 `1–100%` 的比例后，
-   将实际 Process Res 钳制在 `252–1008` 并按 `14` 对齐。
+   模型列表包含三个 MoGe-2 Normal 模型和两个 DA3 模型，默认使用 MoGe-2 ViT-B Normal。DA3 的
+   Resolution Scale 与 Depth Plane 使用相同设计；MoGe-2 改为显示共享的 Resolution Level。
 2. Service 请求 DA3 的 `depth`、`conf` 与 `intrinsics`。仅在开启 `Remove Background` 时运行 BEN2 取得
   前景 Alpha，并在输出前删除包含背景顶点的三角面；关闭时不加载 BEN2。Depth 不做 smoothing，Service
   反投影深度网格，建立源三角面，计算 confidence percentile 与 camera-space edge stretch，并写入
@@ -189,6 +189,11 @@ Cyclic 沿用原 Empty 的设置。序列帧无法被 Blender 整体 Pack，因
   - Smooth 采用 `O Smooth` 的边界检测和模糊法线权重，固定开放边界与薄面，只保留一个
     `Smooth` 程度系数；节点组不再额外执行 Flip Faces；
   - 从修改前的原始表面以 nearest-surface 重新采样 `UVMap`，最终明确写为 Corner 域 `2D Vector`。
+   MoGe-2 使用独立的 `O MoGe2 Surface`。Service 与官方 `infer.py` 一样，先把模型 mask 与
+   `utils3d.np.depth_map_edge(depth, rtol=0.04)` 合并，再由 `utils3d.np.build_mesh_from_map(..., tri=True)`
+   直接生成紧凑三角网格；Blender 不再从规则像素网格重新生面。节点组保留顶层 `Mask`、Resolution、
+   Relax、Smooth、拓扑修复与 UV 重投影，不提供 Confidence、Stretch 或 Remesh 开关；Mask 的 Structure
+   Type 为 `Single`。Remesh 仍是节点处理流程的一部分。
 4. 节点组随插件放在 `O_SnapForm.blend`。Blender 4.5 首次 Append 后复用同一 datablock；
   Blender 5.0 以 linked ID 载入后调用 Pack，后续 Surface 继续复用。插件不会为每个 Surface 新建节点组，
   也不会在添加后重命名 Geometry Nodes 修改器。节点组只启用 `Usage > Modifier`，不 Mark as Asset；
@@ -202,7 +207,8 @@ Cyclic 沿用原 Empty 的设置。序列帧无法被 Blender 整体 Pack，因
   数据后直接创建 Surface 与原图材质，不需要 Open3D 或 OpenCV 进入 Blender Python 环境。
   Surface 保留下列结构化 attributes：
   - 点域：`o_depth`、`o_confidence`、`o_confidence_percentile`、`o_background`；
-   - 面域：`o_face_confidence`、`o_edge_stretch`、`o_background_face`。
+   - 面域：`o_face_confidence`、`o_background_face`；`o_edge_stretch` 位于边域。
+   MoGe-2 的导入网格只保留点域 `o_depth` 与面域 `o_mask`，不伪造 confidence 或 background 属性。
 7. 网格以 `reference_depth` 截面匹配 Image Empty 的显示矩形：该截面的投影四角与 Empty 的可见
   边界重合，主点同时匹配 Empty Offset。网格对象直接沿用源 Empty 的 `matrix_world`，因此位置、旋转、
   缩放及父级关系产生的世界变换保持一致。更近的顶点沿 Empty 局部 `+Z` 方向凸出，三角面正面朝向
@@ -268,7 +274,8 @@ Displacement 设置为 `Displacement Only`，DA3 与 Original 都使用 `Midleve
 `Background Depth`，使两种置换路径保持一致；未移除背景的普通不透明图片因此全部视为前景。
 视频与序列帧的颜色、深度与前景 Image Texture
 通过 image_user 的 Frames / Offset / Start 把第 i 帧映射到场景 `Start+i` 帧，并启用 Auto Refresh。
-颜色贴图到 Principled BSDF 之间使用独立的 Shader Node Group。Surface 使用 `O DA3 Surface Layer`；
+颜色贴图到 Principled BSDF 之间使用独立的 Shader Node Group。DA3 Surface 使用 `O DA3 Surface Layer`，
+MoGe-2 Surface 使用独立的 `O MoGe2 Surface Layer`；
 Depth Plane 使用 `O DA3 Depth Plane Layer` 材质组，并把背景剔除、背景深度与材质置换 Attribute
 读取全部收进组内。两者都接收 `Color`、`Alpha` 与 `Bump Scale`，只有 `O DA3 Depth Plane Layer`
 额外接收 Float 类型的 `Depth`。
--- a/src/snapform/assets/O_SnapForm.blend
+++ b/src/snapform/assets/O_SnapForm.blend
--- a/src/snapform/client.py
+++ b/src/snapform/client.py
@@ -38,6 +38,7 @@ JOB_FOLDER_NAMES = {
    "RECONSTRUCT": "export-reconstruction",
    "IMAGE_DEPTH": "generate-depth-plane",
    "MOGE2_DEPTH": "generate-moge2-depth-plane",
+    "MOGE2_MESH": "generate-moge2-surface",
    "MOGE2_DEBUG": "test-moge2-api",
    "IMAGE_MESH": "generate-image-mesh",
    "REMOVE_BACKGROUND": "remove-background",
@@ -909,6 +910,8 @@ def moge_request(
    start_frame=1,
    device=None,
    model_key=None,
+    debug_export_format="",
+    surface_mesh=False,
 ):
    from .models import BEN2_MODEL, MOGE2_MODEL

@@ -929,6 +932,8 @@ def moge_request(
            else resolution_level
        ),
        "include_points": bool(include_points),
+        "debug_export_format": debug_export_format,
+        "surface_mesh": bool(surface_mesh),
        "video_frames": settings.video_frames,
        "start_frame": start_frame,
        "device": (device or settings.device).lower(),
--- a/src/snapform/image_depth.py
+++ b/src/snapform/image_depth.py
@@ -19,6 +19,7 @@ MATERIAL_DISPLACEMENT_ATTRIBUTE = "o_material_displacement"
 BACKGROUND_DEPTH_ATTRIBUTE = "o_background_depth"
 DEPTH_SCALE_ATTRIBUTE = "o_depth_scale"
 SURFACE_MATERIAL_NODE_GROUP_NAME = "O DA3 Surface Layer"
+MOGE_SURFACE_MATERIAL_NODE_GROUP_NAME = "O MoGe2 Surface Layer"
 DEPTH_PLANE_MATERIAL_NODE_GROUP_NAME = "O DA3 Depth Plane Layer"
 MOGE_MATERIAL_NODE_GROUP_NAME = "O MoGe2 Depth Plane Layer"
 DA3_PATCH_SIZE = 14
@@ -57,9 +58,6 @@ def image_depth_model_items(owner, context):
    return depth_plane_model_enum_items(owner, context)


-_last_depth_plane_model = "MOGE2_VITB_NORMAL"
-
-
 def is_image_empty(candidate):
    return bool(
        candidate
@@ -1006,10 +1004,13 @@ def create_moge_material(
    uv.location = (-900.0, 200.0)
    color = nodes.new("ShaderNodeTexImage")
    color.image = color_image
+    color.interpolation = "Linear"
+    color.extension = "EXTEND"
    color.location = (-650.0, 300.0)
    depth = nodes.new("ShaderNodeTexImage")
    depth.image = depth_image
    depth.interpolation = "Linear"
+    depth.extension = "EXTEND"
    depth.location = (-650.0, -100.0)
    if frame_count > 1:
        configure_image_user(
@@ -1127,11 +1128,11 @@ def create_plane_mesh(source_object, bounds=None, subdivision_level=5):
    return mesh


-def material_node_group(depth_plane=False):
+def material_node_group(depth_plane=False, surface_group_name=None):
    group_name = (
        DEPTH_PLANE_MATERIAL_NODE_GROUP_NAME
        if depth_plane
-        else SURFACE_MATERIAL_NODE_GROUP_NAME
+        else surface_group_name or SURFACE_MATERIAL_NODE_GROUP_NAME
    )
    existing = bpy.data.node_groups.get(group_name)
    if existing is not None:
@@ -1269,7 +1270,7 @@ def material_node_group(depth_plane=False):
    group.description = (
        "Prepare DA3 depth-plane material outputs."
        if depth_plane
-        else "Prepare DA3 surface material outputs."
+        else f"Prepare {group_name} material outputs."
    )
    group.use_fake_user = True
    return group
@@ -1293,6 +1294,7 @@ def create_material(
    depth_frame_offset=0,
    depth_frame_duration=None,
    use_cyclic=None,
+    surface_group_name=None,
 ):
    material = bpy.data.materials.new(name)
    material.use_nodes = True
@@ -1347,7 +1349,10 @@ def create_material(
    alpha = color_texture.outputs["Alpha"]
    material_inputs = nodes.new("ShaderNodeGroup")
    is_depth_plane = displacement_image is not None
-    material_inputs.node_tree = material_node_group(is_depth_plane)
+    material_inputs.node_tree = material_node_group(
+        is_depth_plane,
+        surface_group_name=surface_group_name,
+    )
    material_inputs.name = material_inputs.node_tree.name
    material_inputs.label = material_inputs.name
    material_inputs.location = (-300.0, 350.0)
@@ -2232,35 +2237,21 @@ class ConvertImageToDepthPlane(bpy.types.Operator):
        )

    def invoke(self, context, _event):
-        global _last_depth_plane_model
-
-        if not self.properties.is_property_set("snapform_model"):
-            available_keys = [
-                item[0] for item in image_depth_model_items(self, context)
-            ]
-            if available_keys:
-                self.snapform_model = (
-                    _last_depth_plane_model
-                    if _last_depth_plane_model in available_keys
-                    else available_keys[0]
-                )
        return context.window_manager.invoke_props_dialog(self, width=340)

    def draw(self, context):
        layout = self.layout
        layout.prop(self, "snapform_model")
        if self.snapform_model in {"SMALL", "BASE"}:
-            estimate_box = layout.box()
-            estimate_box.prop(self, "snapform_use_ray_pose")
-            draw_process_resolution(estimate_box, self, context.object.data)
+            layout.prop(self, "snapform_use_ray_pose")
+            draw_process_resolution(layout, self, context.object.data)
        elif self.snapform_model.startswith("MOGE2_"):
            from .client import model_ready

-            moge_box = layout.box()
            if not model_ready(self.snapform_model):
-                moge_box.label(text="MoGe-2 model is not installed", icon="ERROR")
-                moge_box.label(text="Download it in Models")
-            moge_box.prop(self, "snapform_moge_resolution_level")
+                layout.label(text="MoGe-2 model is not installed", icon="ERROR")
+                layout.label(text="Download it in Models")
+            layout.prop(self, "snapform_moge_resolution_level")
        layout.prop(self, "snapform_remove_background")
        layout.prop(self, "snapform_unshaded")
        layout.separator(type="LINE")
@@ -2369,8 +2360,6 @@ class ConvertImageToDepthPlane(bpy.types.Operator):
            if delete_input:
                remove_temporary_input(source_path)
            return {"CANCELLED"}
-        global _last_depth_plane_model
-        _last_depth_plane_model = self.snapform_model
        return {"FINISHED"}


--- a/src/snapform/image_mesh.py
+++ b/src/snapform/image_mesh.py
@@ -3,9 +3,9 @@ from pathlib import Path
 import bpy

 from .image_depth import (
+    MOGE_SURFACE_MATERIAL_NODE_GROUP_NAME,
    create_material,
    draw_process_resolution,
-    image_depth_model_items,
    is_image_empty,
    plane_bounds,
    prepare_image_input,
@@ -16,8 +16,6 @@ from .image_depth import (
 )


-_last_mesh_model = "BASE"
-
 POINT_ATTRIBUTES = {
    "o_depth": ("FLOAT", "point_depth"),
    "o_confidence": ("FLOAT", "point_confidence"),
@@ -32,19 +30,28 @@ FACE_ATTRIBUTES = {
        "FLOAT",
        "face_confidence_percentile",
    ),
-    "o_edge_stretch": ("FLOAT", "face_edge_stretch"),
    "o_background_face": ("BOOLEAN", "face_background"),
 }
+OPTIONAL_FACE_ATTRIBUTES = {
+    "o_mask": ("BOOLEAN", "face_mask"),
+}
 NODE_GROUP_NAME = "O DA3 Surface"
+MOGE_NODE_GROUP_NAME = "O MoGe2 Surface"
 NODE_ASSET_NAME = "O_SnapForm.blend"


-def load_image_mesh_data(npz_path):
+def image_surface_model_items(owner, context):
+    from .models import surface_model_enum_items
+
+    return surface_model_enum_items(owner, context)
+
+
+def load_image_mesh_data(npz_path, moge2=False):
    import numpy as np

    try:
        with np.load(npz_path, allow_pickle=False) as archive:
-            required = (
+            required = [
                "vertices",
                "faces",
                "uv",
@@ -52,12 +59,12 @@ def load_image_mesh_data(npz_path):
                "image_size",
                "reference_depth",
                "point_depth",
-                "point_confidence",
-                "point_confidence_percentile",
-                "point_background",
-                "face_confidence_percentile",
-                "face_edge_stretch",
-                "face_background",
+            ]
+            if moge2:
+                required.append("face_mask")
+            else:
+                required.extend(
+                    ("point_confidence", "point_confidence_percentile", "point_background", "face_confidence_percentile", "face_background")
                )
            missing = [key for key in required if key not in archive.files]
            if missing:
@@ -81,7 +88,16 @@ def load_image_mesh_data(npz_path):
                    **POINT_ATTRIBUTES,
                    **FACE_ATTRIBUTES,
                }.items()
+                if key in archive.files
            }
+            for _name, (_data_type, key) in OPTIONAL_FACE_ATTRIBUTES.items():
+                if key in archive.files:
+                    attributes[key] = np.asarray(archive[key])
+            if "vertex_normal" in archive.files:
+                attributes["vertex_normal"] = np.asarray(
+                    archive["vertex_normal"],
+                    dtype=np.float32,
+                )
    except OSError as error:
        raise ValueError(f"Unable to read surface data: {npz_path}") from error
    if vertices.ndim != 2 or vertices.shape[1] != 3:
@@ -97,10 +113,13 @@ def load_image_mesh_data(npz_path):
    if not reference_depth > 0.0:
        raise ValueError("Surface reference depth must be positive")
    for _name, (_data_type, key) in POINT_ATTRIBUTES.items():
-        if len(attributes[key]) != len(vertices):
+        if key in attributes and len(attributes[key]) != len(vertices):
            raise ValueError(f"Point attribute has the wrong length: {key}")
    for _name, (_data_type, key) in FACE_ATTRIBUTES.items():
-        if len(attributes[key]) != len(faces):
+        if key in attributes and len(attributes[key]) != len(faces):
+            raise ValueError(f"Face attribute has the wrong length: {key}")
+    for _name, (_data_type, key) in OPTIONAL_FACE_ATTRIBUTES.items():
+        if key in attributes and len(attributes[key]) != len(faces):
            raise ValueError(f"Face attribute has the wrong length: {key}")
    return (
        vertices,
@@ -185,7 +204,11 @@ def assign_mesh_attribute(mesh, name, data_type, domain, values):
    attribute.data.foreach_set("value", array.reshape(-1))


-def assign_snapform_attributes(mesh, attributes):
+def assign_snapform_attributes(mesh, attributes, moge2=False):
+    if moge2:
+        assign_mesh_attribute(mesh, "o_depth", "FLOAT", "POINT", attributes["point_depth"])
+        assign_mesh_attribute(mesh, "o_mask", "BOOLEAN", "FACE", attributes["face_mask"])
+        return
    for name, (data_type, key) in POINT_ATTRIBUTES.items():
        assign_mesh_attribute(
            mesh,
@@ -202,6 +225,36 @@ def assign_snapform_attributes(mesh, attributes):
            "FACE",
            attributes[key],
        )
+    for name, (data_type, key) in OPTIONAL_FACE_ATTRIBUTES.items():
+        if key in attributes:
+            assign_mesh_attribute(
+                mesh,
+                name,
+                data_type,
+                "FACE",
+                attributes[key],
+            )
+    assign_edge_depth_difference(mesh, attributes["point_depth"])
+
+
+def assign_edge_depth_difference(mesh, point_depth):
+    import numpy as np
+
+    edge_vertices = np.empty(len(mesh.edges) * 2, dtype=np.int32)
+    mesh.edges.foreach_get("vertices", edge_vertices)
+    edge_vertices = edge_vertices.reshape(-1, 2)
+    depth = np.asarray(point_depth, dtype=np.float32)
+    first = depth[edge_vertices[:, 0]]
+    second = depth[edge_vertices[:, 1]]
+    denominator = np.maximum(np.minimum(np.abs(first), np.abs(second)), 1e-6)
+    difference = np.abs(first - second) / denominator
+    assign_mesh_attribute(
+        mesh,
+        "o_edge_stretch",
+        "FLOAT",
+        "EDGE",
+        difference,
+    )


 def remove_surface_background(vertices, faces, uv, attributes):
@@ -228,12 +281,100 @@ def remove_surface_background(vertices, faces, uv, attributes):
    )


+def prepare_moge_surface_archive(results_path, output_path):
+    import numpy as np
+
+    with np.load(results_path, allow_pickle=False) as archive:
+        required = {"depth", "mask", "points", "intrinsics"}
+        missing = required.difference(archive.files)
+        if missing:
+            raise ValueError(f"MoGe-2 surface data is missing: {', '.join(sorted(missing))}")
+        depth = np.asarray(archive["depth"], dtype=np.float32)
+        mask = np.asarray(archive["mask"], dtype=bool)
+        points = np.asarray(archive["points"], dtype=np.float32)
+        intrinsics = np.asarray(archive["intrinsics"], dtype=np.float32)
+        normal = (
+            np.asarray(archive["normal"], dtype=np.float32)
+            if "normal" in archive.files
+            else None
+        )
+    if depth.shape[0] != 1 or points.shape[:3] != (*depth.shape,):
+        raise ValueError("Convert to Surface requires one matching MoGe-2 point grid")
+    raw_depth = depth[0]
+    finite = np.isfinite(raw_depth) & (raw_depth > 0.0)
+    finite &= np.isfinite(points[0]).all(axis=-1)
+    height, width = raw_depth.shape
+    indices = np.arange(height * width, dtype=np.int32).reshape(height, width)
+    quads = np.stack(
+        (indices[:-1, :-1], indices[1:, :-1], indices[1:, 1:], indices[:-1, 1:]),
+        axis=-1,
+    ).reshape(-1, 4)
+    faces = np.concatenate((quads[:, (0, 1, 2)], quads[:, (0, 2, 3)]), axis=0)
+    faces = faces[finite.reshape(-1)[faces].all(axis=1)]
+    if not len(faces):
+        raise ValueError("MoGe-2 output does not contain a valid surface")
+    flat_points = points[0].reshape(-1, 3)
+    clean_mask = mask[0] & ~depth_discontinuity_mask(raw_depth, 0.04)
+    face_mask = clean_mask.reshape(-1)[faces].all(axis=1)
+    used = np.unique(faces)
+    index_map = np.full(height * width, -1, dtype=np.int32)
+    index_map[used] = np.arange(len(used), dtype=np.int32)
+    u = np.linspace(0.0, 1.0, width, dtype=np.float32)
+    v = np.linspace(1.0, 0.0, height, dtype=np.float32)
+    uv = np.stack(np.meshgrid(u, v), axis=-1).reshape(-1, 2)[used]
+    camera_intrinsics = intrinsics[0].copy()
+    principal_x = float(camera_intrinsics[0, 2])
+    principal_y = float(camera_intrinsics[1, 2])
+    if 0.0 <= principal_x <= 1.0 and 0.0 <= principal_y <= 1.0:
+        camera_intrinsics[0] *= width
+        camera_intrinsics[1] *= height
+        camera_intrinsics[2] = (0.0, 0.0, 1.0)
+    point_depth = raw_depth.reshape(-1)[used]
+    masked_depth = raw_depth[finite & mask[0]]
+    reference_depth = float(np.median(masked_depth if len(masked_depth) else point_depth))
+    np.savez(
+        output_path,
+        vertices=flat_points[used],
+        faces=index_map[faces],
+        uv=uv,
+        intrinsics=camera_intrinsics,
+        image_size=np.array((width, height), dtype=np.int32),
+        reference_depth=np.float32(reference_depth),
+        point_depth=point_depth,
+        vertex_normal=(
+            normal[0].reshape(-1, 3)[used]
+            if normal is not None
+            else np.empty((0, 3), dtype=np.float32)
+        ),
+        face_mask=face_mask,
+    )
+    return output_path
+
+
+def depth_discontinuity_mask(depth, threshold):
+    import numpy as np
+
+    edges = np.zeros(depth.shape, dtype=bool)
+    for axis in (0, 1):
+        first = np.take(depth, range(depth.shape[axis] - 1), axis=axis)
+        second = np.take(depth, range(1, depth.shape[axis]), axis=axis)
+        scale = np.maximum(np.minimum(np.abs(first), np.abs(second)), 1e-6)
+        discontinuity = np.abs(first - second) / scale > threshold
+        lower = [slice(None), slice(None)]
+        upper = [slice(None), slice(None)]
+        lower[axis] = slice(0, -1)
+        upper[axis] = slice(1, None)
+        edges[tuple(lower)] |= discontinuity
+        edges[tuple(upper)] |= discontinuity
+    return edges
+
+
 def node_asset_path():
    return Path(__file__).with_name("assets") / NODE_ASSET_NAME


-def load_geometry_node_group():
-    existing = bpy.data.node_groups.get(NODE_GROUP_NAME)
+def load_geometry_node_group(group_name=NODE_GROUP_NAME):
+    existing = bpy.data.node_groups.get(group_name)
    if existing is not None:
        return existing
    asset_path = node_asset_path()
@@ -243,15 +384,15 @@ def load_geometry_node_group():
    if not use_pack:
        directory = str(asset_path) + "/NodeTree/"
        result = bpy.ops.wm.append(
-            filepath=directory + NODE_GROUP_NAME,
+            filepath=directory + group_name,
            directory=directory,
-            filename=NODE_GROUP_NAME,
+            filename=group_name,
            link=False,
            do_reuse_local_id=True,
        )
        if "FINISHED" not in result:
            raise RuntimeError("Unable to append the Geometry Nodes asset")
-        node_group = bpy.data.node_groups.get(NODE_GROUP_NAME)
+        node_group = bpy.data.node_groups.get(group_name)
        if node_group is None:
            raise RuntimeError("Appended Geometry Nodes asset is unavailable")
        node_group.use_fake_user = True
@@ -260,11 +401,11 @@ def load_geometry_node_group():
        source,
        target,
    ):
-        if NODE_GROUP_NAME not in source.node_groups:
+        if group_name not in source.node_groups:
            raise RuntimeError(
-                f"Geometry Nodes asset is missing {NODE_GROUP_NAME}"
+                f"Geometry Nodes asset is missing {group_name}"
            )
-        target.node_groups = [NODE_GROUP_NAME]
+        target.node_groups = [group_name]
    node_group = target.node_groups[0]
    if use_pack:
        linked_group = node_group
@@ -307,8 +448,13 @@ def material_depth_scale(mesh_object, camera_parameters):
    )


-def add_geometry_nodes_processing(mesh_object, vertices, faces):
-    node_group = load_geometry_node_group()
+def add_geometry_nodes_processing(
+    mesh_object,
+    vertices,
+    faces,
+    group_name=NODE_GROUP_NAME,
+):
+    node_group = load_geometry_node_group(group_name)
    modifier = mesh_object.modifiers.new(name=node_group.name, type="NODES")
    modifier.node_group = node_group
    modifier.show_group_selector = False
@@ -360,7 +506,10 @@ def build_image_mesh(
    create_camera,
    unshaded,
    remove_background=False,
+    moge2=False,
 ):
+    import numpy as np
+
    (
        camera_vertices,
        faces,
@@ -369,7 +518,7 @@ def build_image_mesh(
        image_size,
        reference_depth,
        attributes,
-    ) = load_image_mesh_data(npz_path)
+    ) = load_image_mesh_data(npz_path, moge2=moge2)
    if remove_background:
        camera_vertices, faces, uv, attributes = remove_surface_background(
            camera_vertices,
@@ -388,10 +537,24 @@ def build_image_mesh(
    mesh = bpy.data.meshes.new(source_name)
    mesh.from_pydata(vertices.tolist(), [], faces.tolist())
    assign_vertex_uv(mesh, uv)
-    assign_snapform_attributes(mesh, attributes)
+    assign_snapform_attributes(mesh, attributes, moge2=moge2)
    mesh.update()
    for polygon in mesh.polygons:
        polygon.use_smooth = True
+    if moge2 and len(attributes.get("vertex_normal", ())) == len(vertices):
+        normals = attributes["vertex_normal"].copy()
+        normals[:, 0] /= camera_parameters["scale_x"]
+        normals[:, 1] /= -camera_parameters["scale_y"]
+        normals[:, 2] /= -camera_parameters["scale_z"]
+        lengths = np.linalg.norm(normals, axis=1, keepdims=True)
+        normals /= np.maximum(lengths, 1e-6)
+        set_custom_normals = getattr(
+            mesh,
+            "normals_split_custom_set_from_vertices",
+            None,
+        )
+        if callable(set_custom_normals):
+            set_custom_normals(normals.tolist())
    mesh_object = bpy.data.objects.new(source_name, mesh)
    context.collection.objects.link(mesh_object)
    mesh_object.matrix_world = source_object.matrix_world.copy()
@@ -401,11 +564,19 @@ def build_image_mesh(
        source_object.data,
        depth_scale,
        unshaded=unshaded,
+        surface_group_name=(
+            MOGE_SURFACE_MATERIAL_NODE_GROUP_NAME if moge2 else None
+        ),
    )
    mesh.materials.append(material)
    mesh_object["snapform_reference_depth"] = reference_depth
    mesh_object["snapform_source_type"] = "IMAGE_MESH"
-    add_geometry_nodes_processing(mesh_object, vertices, faces)
+    add_geometry_nodes_processing(
+        mesh_object,
+        vertices,
+        faces,
+        MOGE_NODE_GROUP_NAME if moge2 else NODE_GROUP_NAME,
+    )
    if create_camera:
        create_matching_camera(
            context,
@@ -453,6 +624,10 @@ class BuildImageMesh(bpy.types.Operator):
        default=False,
        options={"HIDDEN", "SKIP_SAVE"},
    )
+    snapform_moge2: bpy.props.BoolProperty(
+        default=False,
+        options={"HIDDEN", "SKIP_SAVE"},
+    )

    def execute(self, context):
        source_object = bpy.data.objects.get(self.snapform_source_object_name)
@@ -468,6 +643,7 @@ class BuildImageMesh(bpy.types.Operator):
                self.snapform_create_camera,
                self.snapform_unshaded,
                self.snapform_remove_background,
+                self.snapform_moge2,
            )
        except (OSError, RuntimeError, ValueError) as error:
            self.report({"ERROR"}, str(error))
@@ -484,7 +660,8 @@ class ConvertImageToMesh(bpy.types.Operator):
    snapform_model: bpy.props.EnumProperty(
        name="Model",
        description="Model used to estimate image depth",
-        items=image_depth_model_items,
+        items=image_surface_model_items,
+        default=1,
    )
    snapform_use_ray_pose: bpy.props.BoolProperty(
        name="Use Ray Pose",
@@ -492,6 +669,13 @@ class ConvertImageToMesh(bpy.types.Operator):
        default=True,
    )
    snapform_process_res_factor: process_resolution_scale_property()
+    snapform_moge_resolution_level: bpy.props.IntProperty(
+        name="Resolution Level",
+        description="MoGe-2 inference resolution level; higher is finer and slower",
+        default=5,
+        min=0,
+        max=9,
+    )
    snapform_create_camera: bpy.props.BoolProperty(
        name="Create Matching Camera",
        description="Create a perspective camera matching the source projection",
@@ -515,7 +699,7 @@ class ConvertImageToMesh(bpy.types.Operator):

    @classmethod
    def poll(cls, context):
-        from .models import MODELS
+        from .models import MODELS, MOGE2_MODELS
        from .operators import active_job
        from .client import service_busy, model_ready

@@ -524,27 +708,21 @@ class ConvertImageToMesh(bpy.types.Operator):
            and video_total_frames(context.object.data) == 1
            and active_job() is None
            and not service_busy()
-            and any(model_ready(model_key) for model_key in MODELS)
+            and any(
+                model_ready(model_key)
+                for model_key in (*MOGE2_MODELS, *MODELS)
+            )
        )

    def invoke(self, context, _event):
-        global _last_mesh_model
-
-        if not self.properties.is_property_set("snapform_model"):
-            available_keys = [
-                item[0] for item in image_depth_model_items(self, context)
-            ]
-            if available_keys:
-                self.snapform_model = (
-                    _last_mesh_model
-                    if _last_mesh_model in available_keys
-                    else available_keys[0]
-                )
        return context.window_manager.invoke_props_dialog(self, width=360)

    def draw(self, context):
        layout = self.layout
        layout.prop(self, "snapform_model")
+        if self.snapform_model.startswith("MOGE2_"):
+            layout.prop(self, "snapform_moge_resolution_level")
+        else:
            layout.prop(self, "snapform_use_ray_pose")
            draw_process_resolution(layout, self, context.object.data)
        layout.prop(self, "snapform_create_camera")
@@ -554,13 +732,15 @@ class ConvertImageToMesh(bpy.types.Operator):
        layout.prop(self, "snapform_keep_original")

    def execute(self, context):
-        from .operators import IMAGE_MESH_JOB
+        from .operators import IMAGE_MESH_JOB, MOGE2_MESH_JOB

        source_object = context.object
        if video_total_frames(source_object.data) != 1:
            self.report({"ERROR"}, "Convert to Surface currently supports still images")
            return {"CANCELLED"}
        try:
+            process_res = 1
+            if not self.snapform_model.startswith("MOGE2_"):
                process_res = scaled_process_resolution(
                    source_object.data,
                    self.snapform_process_res_factor,
@@ -573,12 +753,17 @@ class ConvertImageToMesh(bpy.types.Operator):
            self.report({"ERROR"}, str(error))
            return {"CANCELLED"}
        result = bpy.ops.snapform.run_job(
-            job_type=IMAGE_MESH_JOB,
+            job_type=(
+                MOGE2_MESH_JOB
+                if self.snapform_model.startswith("MOGE2_")
+                else IMAGE_MESH_JOB
+            ),
            snapform_input_path=str(source_path),
            snapform_source_object_name=source_object.name,
            snapform_model_key=self.snapform_model,
            snapform_use_ray_pose=self.snapform_use_ray_pose,
            snapform_process_res=process_res,
+            snapform_moge_resolution_level=self.snapform_moge_resolution_level,
            snapform_delete_source=not self.snapform_keep_original,
            snapform_delete_input=delete_input,
            snapform_mesh_create_camera=self.snapform_create_camera,
@@ -589,6 +774,4 @@ class ConvertImageToMesh(bpy.types.Operator):
            if delete_input:
                remove_temporary_input(source_path)
            return {"CANCELLED"}
-        global _last_mesh_model
-        _last_mesh_model = self.snapform_model
        return {"FINISHED"}
--- a/src/snapform/models.py
+++ b/src/snapform/models.py
@@ -139,6 +139,8 @@ MODEL_NUMBERS = {


 _ENUM_ITEMS_CACHE = {}
+_DEPTH_PLANE_ENUM_ITEMS_CACHE = {}
+_SURFACE_ENUM_ITEMS_CACHE = {}


 def enum_items(supports_gs=None, show_download_status=False):
@@ -213,9 +215,13 @@ def depth_plane_model_enum_items(_owner, _context):
        *MOGE2_MODELS.values(),
        *(model for model in MODELS.values() if model.enabled),
    ]
+    ready = tuple(model_ready(model.key) for model in models)
+    cached = _DEPTH_PLANE_ENUM_ITEMS_CACHE.get(ready)
+    if cached is not None:
+        return cached
    items = []
    for index, model in enumerate(models):
-        label = model.label if model_ready(model.key) else f"{model.label} (Missing)"
+        label = model.label if ready[index] else f"{model.label} (Missing)"
        items.append(
            (
                model.key,
@@ -225,6 +231,22 @@ def depth_plane_model_enum_items(_owner, _context):
            )
        )
    items.append(("ORIGINAL", "Original", "Use the original image as depth", len(items)))
+    _DEPTH_PLANE_ENUM_ITEMS_CACHE[ready] = items
+    return items
+
+
+def surface_model_enum_items(owner, context):
+    depth_items = depth_plane_model_enum_items(owner, context)
+    cache_key = id(depth_items)
+    cached = _SURFACE_ENUM_ITEMS_CACHE.get(cache_key)
+    if cached is not None:
+        return cached
+    items = [
+        item
+        for item in depth_items
+        if item[0] != "ORIGINAL"
+    ]
+    _SURFACE_ENUM_ITEMS_CACHE[cache_key] = items
    return items


--- a/src/snapform/operators.py
+++ b/src/snapform/operators.py
@@ -20,6 +20,7 @@ from .models import (
 RECONSTRUCTION_JOB = "RECONSTRUCT"
 IMAGE_DEPTH_JOB = "IMAGE_DEPTH"
 MOGE2_DEPTH_JOB = "MOGE2_DEPTH"
+MOGE2_MESH_JOB = "MOGE2_MESH"
 MOGE2_DEBUG_JOB = "MOGE2_DEBUG"
 IMAGE_MESH_JOB = "IMAGE_MESH"
 BACKGROUND_REMOVAL_JOB = "REMOVE_BACKGROUND"
@@ -31,6 +32,7 @@ IMAGE_EMPTY_JOBS = frozenset(
    {
        *DA3_IMAGE_JOBS,
        MOGE2_DEPTH_JOB,
+        MOGE2_MESH_JOB,
        IMAGE_BACKGROUND_JOB,
        ORIGINAL_BACKGROUND_JOB,
        ORIGINAL_FRAMES_JOB,
@@ -40,6 +42,7 @@ PRODUCTION_AI_JOBS = frozenset(
    {
        *DA3_IMAGE_JOBS,
        MOGE2_DEPTH_JOB,
+        MOGE2_MESH_JOB,
        IMAGE_BACKGROUND_JOB,
        ORIGINAL_BACKGROUND_JOB,
    }
@@ -163,6 +166,7 @@ class RunJob(bpy.types.Operator):
            (RECONSTRUCTION_JOB, "Export DA3 Result", ""),
            (IMAGE_DEPTH_JOB, "Convert Image to Depth Plane", ""),
            (MOGE2_DEPTH_JOB, "Convert Image with MoGe-2", ""),
+            (MOGE2_MESH_JOB, "Convert Image to Surface with MoGe-2", ""),
            (MOGE2_DEBUG_JOB, "Test MoGe-2 API", ""),
            (IMAGE_MESH_JOB, "Convert Image to Surface", ""),
            (BACKGROUND_REMOVAL_JOB, "Test BEN2 Background Removal", ""),
@@ -250,7 +254,7 @@ class RunJob(bpy.types.Operator):
            export_format = "mini_npz"
        elif self.job_type == IMAGE_MESH_JOB:
            export_format = "mesh_data"
-        elif self.job_type in {MOGE2_DEPTH_JOB, MOGE2_DEBUG_JOB}:
+        elif self.job_type in {MOGE2_DEPTH_JOB, MOGE2_MESH_JOB, MOGE2_DEBUG_JOB}:
            export_format = "moge_npz"

        job = ActiveJob(
@@ -425,8 +429,12 @@ class RunJob(bpy.types.Operator):
                ),
                paths,
            )
-        if self.job_type in {MOGE2_DEPTH_JOB, MOGE2_DEBUG_JOB}:
-            model_key = self.snapform_model_key or MOGE2_MODEL.key
+        if self.job_type in {MOGE2_DEPTH_JOB, MOGE2_MESH_JOB, MOGE2_DEBUG_JOB}:
+            model_key = (
+                settings.moge_model
+                if self.job_type == MOGE2_DEBUG_JOB
+                else self.snapform_model_key or MOGE2_MODEL.key
+            )
            model = get_downloadable_model(model_key)
            if not client.model_ready(model_key):
                raise RuntimeError(f"Download {model.label} first")
@@ -440,17 +448,23 @@ class RunJob(bpy.types.Operator):
                    input_path=input_path,
                    resolution_level=(
                        self.snapform_moge_resolution_level
-                        if self.job_type == MOGE2_DEPTH_JOB
+                        if self.job_type in {MOGE2_DEPTH_JOB, MOGE2_MESH_JOB}
                        else settings.moge_resolution_level
                    ),
-                    include_points=self.job_type == MOGE2_DEBUG_JOB,
+                    include_points=self.job_type in {MOGE2_MESH_JOB, MOGE2_DEBUG_JOB},
                    remove_background=(
-                        self.job_type == MOGE2_DEPTH_JOB
+                        self.job_type in {MOGE2_DEPTH_JOB, MOGE2_MESH_JOB}
                        and remove_background
                    ),
                    start_frame=start_frame,
                    device=self._task_device(settings),
                    model_key=model_key,
+                    debug_export_format=(
+                        settings.moge_export_format
+                        if self.job_type == MOGE2_DEBUG_JOB
+                        else ""
+                    ),
+                    surface_mesh=self.job_type == MOGE2_MESH_JOB,
                ),
                paths,
            )
@@ -531,7 +545,7 @@ class RunJob(bpy.types.Operator):
            return f"Starting {settings.export_format} export"
        if self.job_type == IMAGE_DEPTH_JOB:
            return "Starting depth estimation"
-        if self.job_type in {MOGE2_DEPTH_JOB, MOGE2_DEBUG_JOB}:
+        if self.job_type in {MOGE2_DEPTH_JOB, MOGE2_MESH_JOB, MOGE2_DEBUG_JOB}:
            return "Starting MoGe-2 inference"
        if self.job_type == IMAGE_MESH_JOB:
            return "Starting perspective surface generation"
@@ -728,6 +742,21 @@ class RunJob(bpy.types.Operator):
                self._redraw_ui(context, force=True)
                return {"CANCELLED"}
            status["message"] = "MoGe-2 depth plane is ready"
+        if job.kind == MOGE2_MESH_JOB:
+            try:
+                settings.status = "Building MoGe-2 surface"
+                settings.progress = 0.96
+                self._redraw_ui(context, force=True)
+                self._build_moge2_image_mesh(job)
+            except (OSError, RuntimeError, ValueError) as import_error:
+                settings.status = "MoGe-2 surface import failed"
+                settings.progress = 0.0
+                self.report({"ERROR"}, str(import_error))
+                client.mark_service_job_complete(client.job_id_from_status(job.status_path))
+                close_active_job()
+                self._redraw_ui(context, force=True)
+                return {"CANCELLED"}
+            status["message"] = "MoGe-2 surface is ready"
        if job.kind == IMAGE_MESH_JOB:
            try:
                settings.status = "Building perspective surface"
@@ -801,6 +830,26 @@ class RunJob(bpy.types.Operator):
        if "FINISHED" not in undo_result:
            raise RuntimeError("Unable to create the surface Undo step")

+    def _build_moge2_image_mesh(self, job):
+        result_directory = job.result_directory
+        mesh_path = result_directory / "mesh.npz"
+        if not mesh_path.is_file():
+            raise RuntimeError("MoGe-2 did not export the official surface mesh")
+        result = bpy.ops.snapform.build_image_mesh(
+            snapform_source_object_name=job.source_object_name,
+            snapform_npz_path=str(mesh_path),
+            snapform_delete_source=job.delete_source,
+            snapform_create_camera=job.mesh_create_camera,
+            snapform_unshaded=job.unshaded,
+            snapform_remove_background=False,
+            snapform_moge2=True,
+        )
+        if "FINISHED" not in result:
+            raise RuntimeError("Unable to build the MoGe-2 perspective surface")
+        undo_result = bpy.ops.ed.undo_push(message="Convert to Surface")
+        if "FINISHED" not in undo_result:
+            raise RuntimeError("Unable to create the surface Undo step")
+
    def _build_depth_plane(self, context, job):
        result_directory = job.result_directory
        npz_path = result_directory / "results.npz"
--- a/src/snapform/panel.py
+++ b/src/snapform/panel.py
@@ -252,9 +252,10 @@ class Moge2ApiDebugPanel(bpy.types.Panel):
    def draw(self, context):
        layout = self.layout
        settings = context.scene.snapform_settings
+        layout.prop(settings, "moge_export_format")
        layout.prop(settings, "input_path")
-        layout.prop(settings, "moge_resolution_level")
        layout.prop(settings, "moge_model")
+        layout.prop(settings, "moge_resolution_level")
        run_row = layout.row()
        run_row.enabled = (
            active_job() is None
--- a/src/snapform/properties.py
+++ b/src/snapform/properties.py
@@ -78,6 +78,16 @@ class SnapFormSettings(bpy.types.PropertyGroup):
        items=moge2_model_enum_items,
        default=1,
    )
+    moge_export_format: bpy.props.EnumProperty(
+        name="Output Format",
+        description="Additional MoGe-2 debug output generated with results.npz",
+        items=(
+            ("maps", "Maps", "Export depth, normal, and mask previews"),
+            ("glb", "Textured Surface (GLB)", "Export the official-style textured mesh"),
+            ("ply", "Point Cloud (PLY)", "Export the official-style colored point cloud"),
+        ),
+        default="maps",
+    )
    conf_thresh_percentile: bpy.props.FloatProperty(
        name="Confidence Threshold",
        description="Remove the lowest-confidence percentile of points or surface faces",
--- a/src/snapform/worker.py
+++ b/src/snapform/worker.py
@@ -1514,6 +1514,94 @@ def _save_moge_preview(output_directory, index, depth, normal, mask):
    )


+def export_moge_debug_mesh(
+    output_directory,
+    points,
+    depth,
+    mask,
+    normal,
+    image,
+    export_format,
+):
+    import numpy as np
+    import utils3d
+    from moge.utils.io import save_glb, save_ply
+
+    clean_mask = mask & ~utils3d.np.depth_map_edge(depth, rtol=0.04)
+    uv = utils3d.np.uv_map(*depth.shape)
+    result = utils3d.np.build_mesh_from_map(
+        points,
+        image,
+        uv,
+        normal,
+        mask=clean_mask,
+        tri=True,
+    )
+    faces, vertices, colors, vertex_uvs, vertex_normals = result
+    vertices = vertices * np.array([1.0, -1.0, -1.0], dtype=np.float32)
+    vertex_uvs = vertex_uvs * np.array([1.0, -1.0]) + np.array([0.0, 1.0])
+    vertex_normals = vertex_normals * np.array([1.0, -1.0, -1.0], dtype=np.float32)
+    if export_format == "glb":
+        save_glb(
+            output_directory / "mesh.glb",
+            vertices,
+            faces,
+            vertex_uvs,
+            (image * 255.0).astype(np.uint8),
+            vertex_normals,
+        )
+        return
+    if export_format == "ply":
+        save_ply(
+            output_directory / "pointcloud.ply",
+            vertices,
+            np.zeros((0, 3), dtype=np.int32),
+            colors,
+            vertex_normals,
+        )
+
+
+def export_moge_surface_archive(
+    output_path,
+    points,
+    depth,
+    mask,
+    normal,
+    intrinsics,
+):
+    import numpy as np
+    import utils3d
+
+    height, width = depth.shape
+    clean_mask = mask & ~utils3d.np.depth_map_edge(depth, rtol=0.04)
+    result = utils3d.np.build_mesh_from_map(
+        points,
+        utils3d.np.uv_map(height, width),
+        normal,
+        mask=clean_mask,
+        tri=True,
+    )
+    faces, vertices, vertex_uvs, vertex_normals = result
+    vertex_uvs = vertex_uvs * np.array([1.0, -1.0]) + np.array([0.0, 1.0])
+    pixel_intrinsics = intrinsics.copy()
+    pixel_intrinsics[0] *= width
+    pixel_intrinsics[1] *= height
+    pixel_intrinsics[2] = (0.0, 0.0, 1.0)
+    point_depth = vertices[:, 2].astype(np.float32)
+    np.savez(
+        output_path,
+        vertices=vertices.astype(np.float32),
+        faces=faces.astype(np.int32),
+        uv=vertex_uvs.astype(np.float32),
+        intrinsics=pixel_intrinsics.astype(np.float32),
+        image_size=np.array((width, height), dtype=np.int32),
+        reference_depth=np.float32(np.median(point_depth)),
+        point_depth=point_depth,
+        vertex_normal=vertex_normals.astype(np.float32),
+        face_mask=np.ones(len(faces), dtype=bool),
+    )
+
+
 def run_moge(args, model_cache=None, cancel_check=None):
    import numpy as np
    from PIL import Image
@@ -1550,6 +1638,7 @@ def run_moge(args, model_cache=None, cancel_check=None):
        masks = []
        points = []
        intrinsics = []
+        source_images = []
        for index, image_path in enumerate(images):
            check_cancelled(cancel_check)
            write_status(
@@ -1560,6 +1649,7 @@ def run_moge(args, model_cache=None, cancel_check=None):
            with Image.open(image_path) as opened:
                image = np.asarray(opened.convert("RGB"), dtype=np.float32) / 255.0
            image_tensor = torch.from_numpy(image).permute(2, 0, 1).to(device)
+            source_images.append(image)
            with torch.inference_mode():
                prediction = model.infer(
                    image_tensor,
@@ -1579,6 +1669,8 @@ def run_moge(args, model_cache=None, cancel_check=None):
            intrinsics.append(intrinsic)
            if bool(getattr(args, "include_points", False)):
                points.append(point)
+            debug_format = getattr(args, "debug_export_format", "")
+            if not debug_format or debug_format == "maps":
                _save_moge_preview(
                    output_directory,
                    None if len(images) == 1 else index,
@@ -1598,7 +1690,18 @@ def run_moge(args, model_cache=None, cancel_check=None):
        }
        if points:
            result["points"] = np.stack(points).astype(np.float32)
-        np.savez(output_directory / "results.npz", **result)
+        debug_format = getattr(args, "debug_export_format", "")
+        if debug_format in {"glb", "ply"} and len(images) == 1:
+            write_status(args.status, 0.84, f"Exporting MoGe-2 {debug_format.upper()}")
+            export_moge_debug_mesh(
+                output_directory,
+                points[0],
+                depths[0],
+                masks[0],
+                normals[0],
+                source_images[0],
+                debug_format,
+            )

        remove_background_model_dir = getattr(
            args, "remove_background_model_dir", ""
@@ -1614,7 +1717,14 @@ def run_moge(args, model_cache=None, cancel_check=None):
                    model_cache=model_cache,
                    cancel_check=cancel_check,
                )
-                foreground.save(output_directory / "foreground.png")
+                foreground_path = output_directory / "foreground.png"
+                foreground.save(foreground_path)
+                if points:
+                    foreground_mask = foreground_mesh_mask(
+                        foreground_path,
+                        result["depth"][0].shape,
+                    )
+                    result["mask"][0] &= foreground_mask
            else:
                infer_foreground_frames(
                    images,
@@ -1628,6 +1738,19 @@ def run_moge(args, model_cache=None, cancel_check=None):
                    model_cache=model_cache,
                    cancel_check=cancel_check,
                )
+        if bool(getattr(args, "surface_mesh", False)):
+            if len(images) != 1 or not points:
+                raise RuntimeError("MoGe-2 Surface requires one point map")
+            write_status(args.status, 0.94, "Building official MoGe-2 surface")
+            export_moge_surface_archive(
+                output_directory / "mesh.npz",
+                points[0],
+                result["depth"][0],
+                result["mask"][0],
+                result["normal"][0],
+                result["intrinsics"][0],
+            )
+        np.savez(output_directory / "results.npz", **result)
        copy_frames(images, output_directory)
        write_status(
            args.status,
--- a/tests/test_addon.py
+++ b/tests/test_addon.py
@@ -87,8 +87,6 @@ class AddonTest(unittest.TestCase):
        self.context_menu_draws.clear()
        self.outliner_menu_draws.clear()
        self.addon.operators._active_job = None
-        self.addon.image_depth._last_depth_plane_model = "BASE"
-        self.addon.image_mesh._last_mesh_model = "BASE"
        self.fake_bpy.data = SimpleNamespace(
            objects=SimpleNamespace(get=lambda _name: None),
        )
@@ -732,11 +730,8 @@ class AddonTest(unittest.TestCase):
            ],
        )

-    def test_conversion_dialog_remembers_last_depth_model(self):
+    def test_conversion_dialog_leaves_last_properties_to_blender(self):
        operator = self.addon.ConvertImageToDepthPlane()
-        operator.properties = SimpleNamespace(
-            is_property_set=lambda _name: False,
-        )
        context = SimpleNamespace(
            object=SimpleNamespace(
                data=SimpleNamespace(source="FILE", frame_duration=1)
@@ -749,16 +744,9 @@ class AddonTest(unittest.TestCase):
            ),
        )

-        self.addon.image_depth._last_depth_plane_model = "GIANT_1_1"
-        with patch.object(
-            self.addon.client,
-            "model_ready",
-            side_effect=lambda key: key == "GIANT_1_1",
-        ):
        result = operator.invoke(context, None)

        self.assertIs(result, operator)
-        self.assertNotEqual(getattr(operator, "snapform_model", ""), "GIANT_1_1")

    def test_remove_background_defaults_to_empty_image_frames(self):
        source_object = SimpleNamespace(
@@ -995,10 +983,6 @@ class AddonTest(unittest.TestCase):
        self.assertTrue(calls[0]["snapform_unshaded"])
        self.assertEqual(calls[0]["snapform_depth_data_name"], "img.depth")
        self.assertTrue(calls[0]["snapform_delete_input"])
-        self.assertEqual(
-            self.addon.image_depth._last_depth_plane_model,
-            "GIANT_1_1",
-        )

    def test_packed_image_export_writes_a_png_copy(self):
        class PackedImage:
--- a/tools/build_geometry_nodes_asset.py
+++ b/tools/build_geometry_nodes_asset.py
@@ -4,6 +4,7 @@ import bpy


 GROUP_NAME = "O DA3 Surface"
+MOGE_SURFACE_GROUP_NAME = "O MoGe2 Surface"
 DEPTH_PLANE_GROUP_NAME = "O DA3 Depth Plane"
 MOGE_DEPTH_PLANE_GROUP_NAME = "O MoGe2 Depth Plane"
 MATERIAL_DISPLACEMENT_ATTRIBUTE = "o_material_displacement"
@@ -357,8 +358,13 @@ def add_uv_reprojection(group, source, geometry, location):
    return store_uv.outputs["Geometry"]


-def build_group():
-    group = bpy.data.node_groups.new(GROUP_NAME, "GeometryNodeTree")
+def build_group(
+    group_name=GROUP_NAME,
+    include_confidence=True,
+    include_stretch=True,
+    include_mask=False,
+):
+    group = bpy.data.node_groups.new(group_name, "GeometryNodeTree")
    group.is_modifier = True
    interface = group.interface
    geometry_input_socket = new_socket(
@@ -393,15 +399,27 @@ def build_group():
        "INPUT",
        "NodeSocketFloat",
    )
-    smooth_amount.default_value = 1.0
+    smooth_amount.default_value = 0.5
    smooth_amount.min_value = 0.0
    smooth_amount.max_value = 1.0
    smooth_amount.subtype = "FACTOR"
+    if include_mask:
+        mask_socket = new_socket(
+            interface,
+            "Mask",
+            "INPUT",
+            "NodeSocketBool",
+        )
+        mask_socket.default_value = True
+        mask_socket.structure_type = "SINGLE"

+    culling_panel = None
+    if include_confidence or include_stretch:
        culling_panel = interface.new_panel(
            name="Culling",
            default_closed=False,
        )
+    if include_confidence:
        confidence = new_socket(
            interface,
            "Confidence",
@@ -413,6 +431,7 @@ def build_group():
        confidence.min_value = 0.0
        confidence.max_value = 100.0
        confidence.subtype = "PERCENTAGE"
+    if include_stretch:
        edge = new_socket(
            interface,
            "Stretch",
@@ -444,7 +463,15 @@ def build_group():
    nodes = group.nodes
    links = group.links
    geometry_input = new_group_input(nodes, (-650.0, 50.0), {"Geometry"})
-    confidence_input = new_group_input(nodes, (-1800.0, 800.0), {"Confidence"})
+    confidence_input = None
+    if include_confidence:
+        confidence_input = new_group_input(
+            nodes,
+            (-1800.0, 800.0),
+            {"Confidence"},
+        )
+    stretch_input = None
+    if include_stretch:
        stretch_input = new_group_input(nodes, (-1800.0, 50.0), {"Stretch"})
    remesh_input = new_group_input(
        nodes,
@@ -457,6 +484,8 @@ def build_group():
    group_output = nodes.new("NodeGroupOutput")
    group_output.location = (5650.0, 100.0)

+    geometry = geometry_input.outputs["Geometry"]
+    if include_confidence:
        confidence_attribute = nodes.new("GeometryNodeInputNamedAttribute")
        confidence_attribute.data_type = "FLOAT"
        confidence_attribute.inputs["Name"].default_value = "o_face_confidence"
@@ -484,7 +513,18 @@ def build_group():
            confidence_scale.outputs[0],
            input_socket(confidence_compare, "B", 1),
        )
+        delete_confidence = nodes.new("GeometryNodeDeleteGeometry")
+        delete_confidence.domain = "FACE"
+        delete_confidence.location = (-1050.0, 600.0)
+        links.new(geometry, delete_confidence.inputs["Geometry"])
+        links.new(
+            confidence_compare.outputs["Result"],
+            delete_confidence.inputs["Selection"],
+        )
+        geometry = delete_confidence.outputs["Geometry"]

+    culled_geometry = geometry
+    if include_stretch:
        edge_attribute = nodes.new("GeometryNodeInputNamedAttribute")
        edge_attribute.data_type = "FLOAT"
        edge_attribute.inputs["Name"].default_value = "o_edge_stretch"
@@ -493,53 +533,22 @@ def build_group():
        edge_scale.operation = "MULTIPLY"
        edge_scale.inputs[1].default_value = 0.01
        edge_scale.location = (-1600.0, -50.0)
-    edge_maximum = nodes.new("ShaderNodeMath")
-    edge_maximum.operation = "MAXIMUM"
-    edge_maximum.inputs[1].default_value = 0.0001
-    edge_maximum.location = (-1600.0, 100.0)
-    edge_reciprocal = nodes.new("ShaderNodeMath")
-    edge_reciprocal.operation = "DIVIDE"
-    edge_reciprocal.inputs[0].default_value = 1.0
-    edge_reciprocal.location = (-1400.0, 100.0)
        edge_compare = new_compare(
            nodes,
            "FLOAT",
-        "GREATER_EQUAL",
+            "GREATER_THAN",
            (-1200.0, 350.0),
-        1.0,
+            0.0,
        )
        links.new(stretch_input.outputs["Stretch"], edge_scale.inputs[0])
-    links.new(edge_scale.outputs[0], edge_maximum.inputs[0])
-    links.new(edge_maximum.outputs[0], edge_reciprocal.inputs[1])
        links.new(edge_attribute.outputs["Attribute"], input_socket(edge_compare, "A", 0))
-    links.new(edge_reciprocal.outputs[0], input_socket(edge_compare, "B", 1))
-
-    first_or = nodes.new("FunctionNodeBooleanMath")
-    first_or.operation = "OR"
-    first_or.location = (-950.0, 450.0)
-    links.new(confidence_compare.outputs["Result"], first_or.inputs[0])
-    links.new(edge_compare.outputs["Result"], first_or.inputs[1])
-    full_stretch = new_compare(
-        nodes,
-        "FLOAT",
-        "GREATER_EQUAL",
-        (-950.0, -20.0),
-        0.999999,
-    )
-    links.new(
-        edge_scale.outputs[0],
-        input_socket(full_stretch, "A", 0),
-    )
-    culling_selection = nodes.new("FunctionNodeBooleanMath")
-    culling_selection.operation = "OR"
-    culling_selection.location = (-550.0, 300.0)
-    links.new(first_or.outputs[0], culling_selection.inputs[0])
-    links.new(full_stretch.outputs["Result"], culling_selection.inputs[1])
+        links.new(edge_scale.outputs[0], input_socket(edge_compare, "B", 1))
        delete_geometry = nodes.new("GeometryNodeDeleteGeometry")
-    delete_geometry.domain = "FACE"
+        delete_geometry.domain = "EDGE"
        delete_geometry.location = (-350.0, 100.0)
-    links.new(geometry_input.outputs["Geometry"], delete_geometry.inputs["Geometry"])
-    links.new(culling_selection.outputs[0], delete_geometry.inputs["Selection"])
+        links.new(geometry, delete_geometry.inputs["Geometry"])
+        links.new(edge_compare.outputs["Result"], delete_geometry.inputs["Selection"])
+        culled_geometry = delete_geometry.outputs["Geometry"]

    vertex_neighbors = nodes.new("GeometryNodeInputMeshVertexNeighbors")
    vertex_neighbors.location = (-550.0, -320.0)
@@ -557,7 +566,7 @@ def build_group():
    delete_orphans = nodes.new("GeometryNodeDeleteGeometry")
    delete_orphans.domain = "POINT"
    delete_orphans.location = (-300.0, 100.0)
-    links.new(delete_geometry.outputs["Geometry"], delete_orphans.inputs["Geometry"])
+    links.new(culled_geometry, delete_orphans.inputs["Geometry"])
    links.new(orphan_vertex.outputs["Result"], delete_orphans.inputs["Selection"])

    remeshed = add_remesh_zone(
@@ -622,6 +631,49 @@ def build_group():
    return group


+def configure_moge_surface_group(group):
+    nodes = group.nodes
+    links = group.links
+    mask_input = new_group_input(nodes, (-1800.0, 1050.0), {"Mask"})
+    mask_attribute = nodes.new("GeometryNodeInputNamedAttribute")
+    mask_attribute.data_type = "BOOLEAN"
+    mask_attribute.inputs["Name"].default_value = "o_mask"
+    mask_attribute.location = (-1550.0, 1050.0)
+    invert_mask = nodes.new("FunctionNodeBooleanMath")
+    invert_mask.operation = "NOT"
+    invert_mask.location = (-1350.0, 1050.0)
+    use_mask = nodes.new("FunctionNodeBooleanMath")
+    use_mask.operation = "AND"
+    use_mask.location = (-1150.0, 1050.0)
+    links.new(mask_attribute.outputs["Attribute"], invert_mask.inputs[0])
+    links.new(mask_input.outputs["Mask"], use_mask.inputs[0])
+    links.new(invert_mask.outputs[0], use_mask.inputs[1])
+    first_processing = min(
+        (
+            node
+            for node in nodes
+            if node.bl_idname == "GeometryNodeDeleteGeometry"
+        ),
+        key=lambda node: node.location.x,
+    )
+    geometry_input = first_processing.inputs["Geometry"]
+    previous_link = geometry_input.links[0]
+    previous_geometry = previous_link.from_socket
+    links.remove(previous_link)
+    delete_mask = nodes.new("GeometryNodeDeleteGeometry")
+    delete_mask.domain = "FACE"
+    delete_mask.location = (-750.0, 650.0)
+    links.new(previous_geometry, delete_mask.inputs["Geometry"])
+    links.new(use_mask.outputs[0], delete_mask.inputs["Selection"])
+    links.new(delete_mask.outputs["Geometry"], geometry_input)
+    group.description = (
+        "Cull MoGe-2 stretched faces, repair topology, preserve boundaries, "
+        "and reproject UVs."
+    )
+    group.use_fake_user = True
+    return group
+
+
 def build_sequence_frame(nodes, links, scene_frame, frames_input, y):
    relative_frame = nodes.new("ShaderNodeMath")
    relative_frame.operation = "SUBTRACT"
@@ -1246,6 +1298,13 @@ def main():
    bpy.ops.wm.read_factory_settings(use_empty=True)
    bpy.context.preferences.filepaths.save_version = 0
    build_group()
+    moge_surface_group = build_group(
+        MOGE_SURFACE_GROUP_NAME,
+        include_confidence=False,
+        include_stretch=False,
+        include_mask=True,
+    )
+    configure_moge_surface_group(moge_surface_group)
    build_depth_plane_group()
    build_moge_depth_plane_group()
    OUTPUT_PATH.parent.mkdir(parents=True, exist_ok=True)
--- a/tools/validate_depth_plane_asset.py
+++ b/tools/validate_depth_plane_asset.py
@@ -20,6 +20,13 @@ from snapform.image_mesh import load_geometry_node_group
 def main():
    bpy.ops.wm.read_factory_settings(use_empty=True)
    group = load_depth_plane_node_group()
+    geometry_textures = [
+        node
+        for node in group.nodes
+        if node.bl_idname == "GeometryNodeImageTexture"
+    ]
+    assert geometry_textures
+    assert all(node.extension == "EXTEND" for node in geometry_textures)
    background_depth = next(
        item
        for item in group.interface.items_tree
@@ -212,6 +219,11 @@ def main():
        node.bl_idname in forbidden_outer_nodes
        for node in material.node_tree.nodes
    )
+    assert all(
+        node.extension == "EXTEND"
+        for node in material.node_tree.nodes
+        if node.bl_idname == "ShaderNodeTexImage"
+    )
    surface_material = create_material("Surface", foreground, 1.0)
    surface_group = bpy.data.node_groups[SURFACE_MATERIAL_NODE_GROUP_NAME]
    surface_group_nodes = [
@@ -251,6 +263,11 @@ def main():
        node.bl_idname in forbidden_outer_nodes
        for node in surface_material.node_tree.nodes
    )
+    assert all(
+        node.extension == "EXTEND"
+        for node in surface_material.node_tree.nodes
+        if node.bl_idname == "ShaderNodeTexImage"
+    )
    modifier = add_depth_plane_processing(
        obj,
        depth,
--- a/tools/validate_geometry_nodes_asset.py
+++ b/tools/validate_geometry_nodes_asset.py
@@ -22,9 +22,8 @@ def set_input(modifier, group, name, value):


 def add_face_attributes(mesh, count, background_values=None):
-    for name in ("o_face_confidence", "o_edge_stretch"):
-        attribute = mesh.attributes.new(name, "FLOAT", "FACE")
-        attribute.data.foreach_set("value", [1.0] * count)
+    confidence = mesh.attributes.new("o_face_confidence", "FLOAT", "FACE")
+    confidence.data.foreach_set("value", [1.0] * count)
    background = mesh.attributes.new("o_background_face", "BOOLEAN", "FACE")
    background.data.foreach_set(
        "value",
@@ -56,7 +55,7 @@ def main():
        visible_outputs = [
            output for output in node.outputs if output.name and not output.hide
        ]
-        assert 1 <= len(visible_outputs) <= 3
+        assert 1 <= len(visible_outputs) <= 4
    uv_stores = [
        node
        for node in group.nodes
@@ -79,6 +78,14 @@ def main():
    assert [item.name for item in panels["Options"].interface_items] == [
        "Fix Topology",
    ]
+    smooth = next(
+        item
+        for item in group.interface.items_tree
+        if item.item_type == "SOCKET"
+        and item.in_out == "INPUT"
+        and item.name == "Smooth"
+    )
+    assert smooth.default_value == 0.5
    for name, default_value in (("Confidence", 0.0), ("Stretch", 10.0)):
        item = next(
            item
@@ -98,6 +105,8 @@ def main():
        [],
        [(0, 2, 1), (1, 2, 3)],
    )
+    edge_stretch = mesh.attributes.new("o_edge_stretch", "FLOAT", "EDGE")
+    edge_stretch.data.foreach_set("value", [0.05] * len(mesh.edges))
    uv = mesh.uv_layers.new(name="UVMap")
    uv.data.foreach_set("uv", [0, 0, 0, 1, 1, 0, 1, 0, 0, 1, 1, 1])
    add_face_attributes(mesh, 2, [False, True])
@@ -121,13 +130,13 @@ def main():
    finally:
        evaluated.to_mesh_clear()

-    set_input(modifier, group, "Stretch", 100.0)
+    set_input(modifier, group, "Stretch", 0.0)
    stretch_item = next(
        item
        for item in group.interface.items_tree
        if item.item_type == "SOCKET" and item.name == "Stretch"
    )
-    assert modifier[stretch_item.identifier] == 100.0
+    assert modifier[stretch_item.identifier] == 0.0
    obj.update_tag(refresh={"DATA"})
    bpy.context.view_layer.update()
    culled = obj.evaluated_get(bpy.context.evaluated_depsgraph_get())
--- a/tools/validate_moge_depth_plane.py
+++ b/tools/validate_moge_depth_plane.py
@@ -72,6 +72,21 @@ def main():
    material_group = bpy.data.node_groups[MOGE_MATERIAL_NODE_GROUP_NAME]
    assert not any(node.bl_idname == "ShaderNodeNormalMap" for node in material_group.nodes)
    assert any(node.bl_idname == "ShaderNodeBump" for node in material_group.nodes)
+    material = plane.data.materials[0]
+    image_textures = [
+        node
+        for node in material.node_tree.nodes
+        if node.bl_idname == "ShaderNodeTexImage"
+    ]
+    assert image_textures
+    assert all(node.extension == "EXTEND" for node in image_textures)
+    geometry_textures = [
+        node
+        for node in geometry_group.nodes
+        if node.bl_idname == "GeometryNodeImageTexture"
+    ]
+    assert geometry_textures
+    assert all(node.extension == "EXTEND" for node in geometry_textures)
    print("MoGe-2 depth plane validation passed")


--- a/tools/validate_moge_surface.py
+++ b/tools/validate_moge_surface.py
@@ -0,0 +1,176 @@
+import sys
+import tempfile
+from pathlib import Path
+
+import bpy
+import numpy as np
+
+
+PROJECT_ROOT = Path(__file__).parents[1]
+sys.path.insert(0, str(PROJECT_ROOT / "src"))
+
+from snapform.image_depth import MOGE_SURFACE_MATERIAL_NODE_GROUP_NAME
+from snapform.image_mesh import (
+    MOGE_NODE_GROUP_NAME,
+    build_image_mesh,
+    prepare_moge_surface_archive,
+)
+
+
+def main():
+    bpy.ops.wm.read_factory_settings(use_empty=True)
+    arguments = sys.argv[sys.argv.index("--") + 1:] if "--" in sys.argv else []
+    if arguments:
+        validate_real_result(Path(arguments[0]), Path(arguments[1]))
+        return
+    height, width = 48, 64
+    u, v = np.meshgrid(
+        np.linspace(-0.5, 0.5, width, dtype=np.float32),
+        np.linspace(-0.375, 0.375, height, dtype=np.float32),
+    )
+    depth = np.full((1, height, width), 2.0, dtype=np.float32)
+    points = np.stack((u * 2.0, v * 2.0, depth[0]), axis=-1)[None]
+    mask = np.ones((1, height, width), dtype=bool)
+    mask[:, :2] = False
+    intrinsics = np.array(
+        [[[1.0, 0.0, 0.5], [0.0, 1.0, 0.5], [0.0, 0.0, 1.0]]],
+        dtype=np.float32,
+    )
+    image = bpy.data.images.new("MoGe Surface Test", width=width, height=height)
+    source = bpy.data.objects.new("MoGe Surface Test", None)
+    source.empty_display_type = "IMAGE"
+    source.data = image
+    bpy.context.collection.objects.link(source)
+    with tempfile.TemporaryDirectory() as temporary:
+        directory = Path(temporary)
+        results_path = directory / "results.npz"
+        mesh_path = directory / "mesh.npz"
+        np.savez(
+            results_path,
+            depth=depth,
+            mask=mask,
+            points=points,
+            intrinsics=intrinsics,
+        )
+        prepare_moge_surface_archive(results_path, mesh_path)
+        surface = build_image_mesh(
+            bpy.context,
+            source,
+            mesh_path,
+            False,
+            True,
+            False,
+            False,
+            True,
+        )
+    assert len(surface.data.vertices) > 0
+    assert surface.modifiers[0].node_group.name == MOGE_NODE_GROUP_NAME
+    inputs = {
+        item.name
+        for item in surface.modifiers[0].node_group.interface.items_tree
+        if getattr(item, "in_out", None) == "INPUT"
+    }
+    assert "Stretch" not in inputs
+    assert "Mask" in inputs
+    assert "Confidence" not in inputs
+    assert "o_mask" in surface.data.attributes
+    assert surface.data.attributes["o_mask"].domain == "FACE"
+    attribute_names = set(surface.data.attributes.keys())
+    assert attribute_names >= {"o_depth", "o_mask"}
+    assert "o_edge_stretch" not in attribute_names
+    assert not {
+        "o_confidence",
+        "o_confidence_percentile",
+        "o_face_confidence",
+        "o_background",
+        "o_background_face",
+    }.intersection(attribute_names)
+    mask_socket = next(
+        item
+        for item in surface.modifiers[0].node_group.interface.items_tree
+        if getattr(item, "in_out", None) == "INPUT" and item.name == "Mask"
+    )
+    assert mask_socket.default_value is True
+    assert mask_socket.structure_type == "SINGLE"
+    assert not mask_socket.parent.name
+    assert not any(
+        item.item_type == "PANEL" and item.name == "Culling"
+        for item in surface.modifiers[0].node_group.interface.items_tree
+    )
+    smooth_socket = next(
+        item
+        for item in surface.modifiers[0].node_group.interface.items_tree
+        if getattr(item, "in_out", None) == "INPUT" and item.name == "Smooth"
+    )
+    assert smooth_socket.default_value == 0.5
+    assert not any(
+        node.bl_idname == "GeometryNodeInputNamedAttribute"
+        and node.inputs["Name"].default_value == "o_face_confidence"
+        for node in surface.modifiers[0].node_group.nodes
+    )
+    assert MOGE_SURFACE_MATERIAL_NODE_GROUP_NAME in bpy.data.node_groups
+    material_group = bpy.data.node_groups[MOGE_SURFACE_MATERIAL_NODE_GROUP_NAME]
+    assert any(node.bl_idname == "ShaderNodeBump" for node in material_group.nodes)
+    assert any(child.type == "CAMERA" for child in surface.children)
+    print("MoGe-2 surface validation passed")
+
+
+def validate_real_result(results_path, image_path):
+    image = bpy.data.images.load(str(image_path), check_existing=False)
+    source = bpy.data.objects.new("MoGe Real Surface", None)
+    source.empty_display_type = "IMAGE"
+    source.data = image
+    bpy.context.collection.objects.link(source)
+    mesh_path = results_path.with_name("mesh.npz")
+    if not mesh_path.is_file():
+        prepare_moge_surface_archive(results_path, mesh_path)
+    surface = build_image_mesh(
+        bpy.context,
+        source,
+        mesh_path,
+        False,
+        True,
+        False,
+        False,
+        True,
+    )
+    assert len(surface.data.vertices) > 100
+    assert len(surface.data.polygons) > 100
+    assert surface.modifiers[0].node_group.name == MOGE_NODE_GROUP_NAME
+    assert MOGE_SURFACE_MATERIAL_NODE_GROUP_NAME in bpy.data.node_groups
+    assert surface.data.has_custom_normals
+    normals = np.empty(len(surface.data.corner_normals) * 3, dtype=np.float32)
+    surface.data.corner_normals.foreach_get("vector", normals)
+    normals = normals.reshape(-1, 3)
+    assert np.isfinite(normals).all()
+    assert np.min(np.linalg.norm(normals, axis=1)) > 0.99
+    material_group = bpy.data.node_groups[MOGE_SURFACE_MATERIAL_NODE_GROUP_NAME]
+    assert any(node.bl_idname == "ShaderNodeBump" for node in material_group.nodes)
+    attribute_names = set(surface.data.attributes.keys())
+    assert {"o_depth", "o_mask"}.issubset(attribute_names)
+    assert "o_edge_stretch" not in attribute_names
+    assert not {
+        "o_confidence",
+        "o_confidence_percentile",
+        "o_face_confidence",
+        "o_background",
+        "o_background_face",
+    }.intersection(attribute_names)
+    camera = next(child for child in surface.children if child.type == "CAMERA")
+    dimensions = surface.dimensions
+    assert max(dimensions) < 10.0
+    assert min(dimensions.x, dimensions.y) > 0.1
+    assert 1.0 <= camera.data.lens <= 500.0
+    assert abs(camera.data.shift_x) <= 1.0
+    assert abs(camera.data.shift_y) <= 1.0
+    print(
+        f"Real MoGe-2 surface validation passed: "
+        f"{len(surface.data.vertices)} vertices, {len(surface.data.polygons)} faces, "
+        f"dimensions={tuple(round(value, 4) for value in dimensions)}, "
+        f"lens={camera.data.lens:.3f}, shift=({camera.data.shift_x:.4f}, "
+        f"{camera.data.shift_y:.4f})"
+    )
+
+
+if __name__ == "__main__":
+    main()