add octree depth controls fix 2

app.py

@@ -67,10 +67,10 @@ rmbg_net = BriaRMBG.from_pretrained(rmbg_path).to(device)
 rmbg_net.eval()
 
 
-def generate(image_path, face_number=50000, guidance_scale=5.0, num_steps=25):
-# def generate(image_path, face_number=50000, guidance_scale=5.0, num_steps=25, octree_depth=9):
-    # print(f"[INPUT] face_number={face_number}, guidance_scale={guidance_scale}, num_steps={num_steps}, octree_depth={octree_depth}")# 👈 добавлено
-    print(f"[INPUT] face_number={face_number}, guidance_scale={guidance_scale}, num_steps={num_steps}")# 👈 добавлено
+# def generate(image_path, face_number=50000, guidance_scale=5.0, num_steps=25):
+def generate(image_path, face_number=50000, guidance_scale=5.0, num_steps=25, octree_depth=9):
+    print(f"[INPUT] face_number={face_number}, guidance_scale={guidance_scale}, num_steps={num_steps}, octree_depth={octree_depth}")# 👈 добавлено_et
+    # print(f"[INPUT] face_number={face_number}, guidance_scale={guidance_scale}, num_steps={num_steps}")# 👈 добавлено_et
     print("[API CALL] image_path received:", image_path)
     print("[API CALL] File exists:", os.path.exists(image_path))
 
@@ -87,7 +87,7 @@ def generate(image_path, face_number=50000, guidance_scale=5.0, num_steps=25):
             num_inference_steps=int(num_steps),
             guidance_scale=float(guidance_scale),
             faces=int(face_number),
-            # octree_depth=int(octree_depth), # 👈 добавлено
+            octree_depth=int(octree_depth), # 👈 добавлено_et
         )
 
         if mesh is None or mesh.vertices.shape[0] == 0 or mesh.faces.shape[0] == 0:
@@ -129,14 +129,14 @@ def generate(image_path, face_number=50000, guidance_scale=5.0, num_steps=25):
 # Интерфейс Gradio
 demo = gr.Interface(
     fn=generate,
-    inputs=gr.Image(type="filepath", label="Upload image"),
-    # inputs=[
-    #     gr.Image(type="filepath", label="Upload image"),
-    #     gr.Slider(10000, 150000, step=10000, value=50000, label="Face count"),
-    #     gr.Slider(1.0, 10.0, step=0.5, value=5.0, label="Guidance Scale"),
-    #     gr.Slider(10, 100, step=5, value=25, label="Steps"),
-    #     gr.Slider(6, 10, step=1, value=9, label="Octree Depth"),
-    # ], # 👈 добавлено
+    # inputs=gr.Image(type="filepath", label="Upload image"),
+    inputs=[
+        gr.Image(type="filepath", label="Upload image"),
+        gr.Slider(10000, 150000, step=10000, value=50000, label="Face count"),
+        gr.Slider(1.0, 10.0, step=0.5, value=5.0, label="Guidance Scale"),
+        gr.Slider(10, 100, step=5, value=25, label="Steps"),
+        gr.Slider(6, 9, step=1, value=9, label="Octree Depth"),
+    ], # 👈 добавлено
     outputs=gr.File(label="Download .glb"),
     title="TripoSG Image to 3D",
     description="Upload an image to generate a 3D model (.glb)",

inference_triposg.py

@@ -19,32 +19,6 @@ from briarmbg import BriaRMBG
 import pymeshlab
 
 
-# @torch.no_grad()
-# def run_triposg(
-#     pipe: Any,
-#     image_input: Union[str, Image.Image],
-#     rmbg_net: Any,
-#     seed: int,
-#     num_inference_steps: int = 50,
-#     guidance_scale: float = 7.0,
-#     faces: int = -1,
-# ) -> trimesh.Scene:
-
-#     img_pil = prepare_image(image_input, bg_color=np.array([1.0, 1.0, 1.0]), rmbg_net=rmbg_net)
-
-#     outputs = pipe(
-#         image=img_pil,
-#         generator=torch.Generator(device=pipe.device).manual_seed(seed),
-#         num_inference_steps=num_inference_steps,
-#         guidance_scale=guidance_scale,
-#     ).samples[0]
-#     mesh = trimesh.Trimesh(outputs[0].astype(np.float32), np.ascontiguousarray(outputs[1]))
-
-#     if faces > 0:
-#         mesh = simplify_mesh(mesh, faces)
-
-#     return mesh
-
 @torch.no_grad()
 def run_triposg(
     pipe: Any,
@@ -54,7 +28,7 @@ def run_triposg(
     num_inference_steps: int = 50,
     guidance_scale: float = 7.0,
     faces: int = -1,
-    # octree_depth: int = 9, # 👈 добавлено
+    octree_depth: int = 9, # 👈 добавлено_et
 ) -> trimesh.Scene:
     print("[DEBUG] Preparing image...")
     img_pil = prepare_image(image_input, bg_color=np.array([1.0, 1.0, 1.0]), rmbg_net=rmbg_net)
@@ -65,7 +39,7 @@ def run_triposg(
         generator=torch.Generator(device=pipe.device).manual_seed(seed),
         num_inference_steps=num_inference_steps,
         guidance_scale=guidance_scale,
-        # flash_octree_depth=octree_depth, # 👈 добавлено
+        flash_octree_depth=octree_depth, # 👈 добавлено_et
     ).samples[0]
 
     print("[DEBUG] TripoSG output keys:", type(outputs), outputs[0].shape, outputs[1].shape)
@@ -75,7 +49,7 @@ def run_triposg(
 
     if faces > 0:
         print(f"[DEBUG] Simplifying mesh to {faces} faces")
-        # mesh = simplify_mesh(mesh, faces)
+        mesh = simplify_mesh(mesh, faces) # 👈 добавлено_et
 
     return mesh
 
@@ -91,6 +65,7 @@ def pymesh_to_trimesh(mesh):
     faces = mesh.face_matrix()#.tolist()
     return trimesh.Trimesh(vertices=verts, faces=faces)  #, vID, fID
 
+# old version
 # def simplify_mesh(mesh: trimesh.Trimesh, n_faces):
 #     if mesh.faces.shape[0] > n_faces:
 #         ms = mesh_to_pymesh(mesh.vertices, mesh.faces)
@@ -100,17 +75,38 @@ def pymesh_to_trimesh(mesh):
 #     else:
 #         return mesh
 
+# new version
+# def simplify_mesh(mesh: trimesh.Trimesh, n_faces):
+#     if mesh.faces.shape[0] > n_faces:
+#         ms = mesh_to_pymesh(mesh.vertices, mesh.faces)
+#         ms.meshing_merge_close_vertices()
+#         ms.meshing_decimation_quadric_edge_collapse(targetfacenum=n_faces)
+#         simplified = ms.current_mesh()
+#         if simplified is None or simplified.face_number() == 0:
+#             return None
+#         return pymesh_to_trimesh(simplified)
+#     return mesh
+
+# new version demo
 def simplify_mesh(mesh: trimesh.Trimesh, n_faces):
-    if mesh.faces.shape[0] > n_faces:
+    original_faces = mesh.faces.shape[0]  # 👈 сохраняем исходное количество
+
+    if original_faces > n_faces:
         ms = mesh_to_pymesh(mesh.vertices, mesh.faces)
         ms.meshing_merge_close_vertices()
         ms.meshing_decimation_quadric_edge_collapse(targetfacenum=n_faces)
         simplified = ms.current_mesh()
         if simplified is None or simplified.face_number() == 0:
             return None
+
+        simplified_faces = simplified.face_number()
+        print(f"[DEBUG] Simplified mesh: {original_faces} → {simplified_faces} faces")  # 👈 лог здесь
+
         return pymesh_to_trimesh(simplified)
+
     return mesh
 
+
 if __name__ == "__main__":
     device = "cuda"
     dtype = torch.float16