H5ad generation

.gitignore

@@ -0,0 +1,2 @@
+*_marker_genes.txt
+*_scaled.csv

embryo-backend/Data/glbTrans.py

@@ -19,7 +19,7 @@ assert os.path.exists(gene_list_file), f"❌ 缺少基因名文件 {gene_list_fi
 all_gene_names = pd.read_csv(gene_list_file, header=None)[0].tolist()
 gene_names = all_gene_names[:n_genes]
 
-# ==== 真实 marker 基因 ====
+# ==== marker 基因 ====
 true_markers = {
     "Ectoderm": ["SOX2", "PAX6", "NES", "TUBB3", "OTX2"],
     "Mesoderm": ["TBXT", "MESP1", "HAND1", "GATA4", "PDGFRA"],
@@ -30,6 +30,7 @@ true_markers = {
 
 # ==== 采样函数 ====
 def sample_mesh(mesh, label, n_samples):
+    """采样点云 + 颜色 + 标签"""
     if use_surface_sampling:
         points, face_idx = trimesh.sample.sample_surface(mesh, n_samples)
         if hasattr(mesh.visual, "vertex_colors") and len(mesh.visual.vertex_colors) == len(mesh.vertices):
@@ -63,7 +64,7 @@ for glb_name in glb_files:
     scene = trimesh.load(glb_path)
     if isinstance(scene, trimesh.Scene):
         for geom in scene.geometry.values():
-            all_bounds.append(geom.bounds)  # (min,max)
+            all_bounds.append(geom.bounds)
     else:
         all_bounds.append(scene.bounds)
 
@@ -73,7 +74,7 @@ global_max = np.max(all_bounds[:,1,:], axis=0)
 global_size = np.max(global_max - global_min)
 print(f"🌍 统一缩放基准: global_size={global_size}")
 
-# ==== Step 2: 处理单个 GLB ====
+# ==== Step 2: 处理单个 GLB（应用全局变换） ====
 def process_glb(glb_path, sample_name):
     scene = trimesh.load(glb_path)
     all_points, all_colors, all_labels = [], [], []
@@ -83,7 +84,21 @@ def process_glb(glb_path, sample_name):
         for name, geom in scene.geometry.items():
             ratio = len(geom.vertices) / total_points
             n_samples = max(1, int(N_total * ratio))
-            p, c, l = sample_mesh(geom, name, n_samples)
+
+            # ✅ 保护性获取变换矩阵
+            try:
+                transform = scene.graph.get(name)[0]
+            except Exception:
+                print(f"⚠️ {sample_name}: 子网格 {name} 没有变换路径，使用单位矩阵")
+                transform = np.eye(4)
+
+            # ✅ 应用全局变换
+            verts_world = trimesh.transform_points(geom.vertices, transform)
+
+            # ✅ 生成全局 mesh 进行采样
+            mesh_world = trimesh.Trimesh(vertices=verts_world, faces=geom.faces, process=False)
+            p, c, l = sample_mesh(mesh_world, name, n_samples)
+
             all_points.append(p)
             all_colors.append(c)
             all_labels.append(l)
@@ -101,26 +116,21 @@ def process_glb(glb_path, sample_name):
         idx = np.random.choice(len(points), N_total, replace=False)
         points, colors, labels = points[idx], colors[idx], labels[idx]
 
-    # ✅ 平移到中心点
+    # ✅ 一次性中心化 + 统一缩放
     center = points.mean(axis=0)
     points_centered = points - center
-
-    # ✅ 统一缩放（保持原始比例）
     points_scaled = points_centered / global_size
 
-    # ✅ 如需修正坐标系方向，可启用以下行（示例：交换Y和Z）
-    # points_scaled = points_scaled[:, [0, 2, 1]]
-    # points_scaled[:, 1] *= -1
-
     df = pd.DataFrame(points_scaled, columns=["x","y","z"])
     df["r"], df["g"], df["b"] = colors[:,0], colors[:,1], colors[:,2]
     df["label"] = labels
 
     csv_file = os.path.join(script_dir, f"{sample_name}_point_cloud_30000_centered_scaled.csv")
     df.to_csv(csv_file, index=False)
-    print(f"✅ 已导出 {csv_file} (中心化 & 统一大小, 保持比例)")
+    print(f"✅ 已导出 {csv_file} (修正组织错位, 中心化 & 统一比例)")
     return df
 
+
 # ==== Step 3: 生成 h5ad ====
 def create_h5ad(df, sample_name):
     points = df[['x','y','z']].to_numpy()

pyproject.toml

@@ -9,4 +9,5 @@ dependencies = [
     "flask>=3.1.1",
     "flask-cors>=6.0.1",
     "pandas>=2.3.1",
+    "trimesh>=4.7.1",
 ]

uv.lock

@@ -100,6 +100,7 @@ dependencies = [
     { name = "flask" },
     { name = "flask-cors" },
     { name = "pandas" },
+    { name = "trimesh" },
 ]
 
 [package.metadata]
@@ -108,6 +109,7 @@ requires-dist = [
     { name = "flask", specifier = ">=3.1.1" },
     { name = "flask-cors", specifier = ">=6.0.1" },
     { name = "pandas", specifier = ">=2.3.1" },
+    { name = "trimesh", specifier = ">=4.7.1" },
 ]
 
 [[package]]
@@ -421,6 +423,18 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/b7/ce/149a00dd41f10bc29e5921b496af8b574d8413afcd5e30dfa0ed46c2cc5e/six-1.17.0-py2.py3-none-any.whl", hash = "sha256:4721f391ed90541fddacab5acf947aa0d3dc7d27b2e1e8eda2be8970586c3274", size = 11050, upload-time = "2024-12-04T17:35:26.475Z" },
 ]
 
+[[package]]
+name = "trimesh"
+version = "4.7.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "numpy" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/95/05/50656189ebd0563b2130ddd9b609b1db99eb241687dac0d2585882c35c33/trimesh-4.7.1.tar.gz", hash = "sha256:3863c2b2281fc7a99bf0b5de4a0011229bde4663babc0c1b53a1f855149ec898", size = 800778, upload-time = "2025-07-16T20:24:50.248Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/c6/64/99e785dd4bb5796881b5686b0869a6ebf75b1b7d00932af577eb343c4f72/trimesh-4.7.1-py3-none-any.whl", hash = "sha256:338c938ae78ad5b4d08dd6ceaa739498a47627bf0073147ee9a384ddd7435267", size = 709034, upload-time = "2025-07-16T20:24:47.256Z" },
+]
+
 [[package]]
 name = "typing-extensions"
 version = "4.14.1"