solve merge

.gitignore

@@ -3,7 +3,7 @@
 __pycache__/
 *.py[cod]
 *$py.class
-
+temp*
 # C extensions
 *.so
 

data_generator.py

@@ -6,8 +6,9 @@ import bpy
 import numpy as np
 import mathutils
 import requests
-from blender.blender_util import BlenderUtils
-from blender.view_sample_util import ViewSampleUtil
+from utils.blender_util import BlenderUtils
+from utils.view_sample_util import ViewSampleUtil
+from utils.material_util import MaterialUtil
 
 class DataGenerator:
     def __init__(self, config):
@@ -17,6 +18,9 @@ class DataGenerator:
         self.random_config = config["runner"]["generate"]["random_config"]
         self.light_and_camera_config = config["runner"]["generate"]["light_and_camera_config"]
         self.obj_dir = config["runner"]["generate"]["object_dir"]
+        self.use_list = config["runner"]["generate"]["use_list"]
+        self.object_list_path = config["runner"]["generate"]["object_list_path"]
+        
         self.max_views = config["runner"]["generate"]["max_views"]
         self.min_views = config["runner"]["generate"]["min_views"]
         self.min_diag = config["runner"]["generate"]["min_diag"]
@@ -29,7 +33,18 @@ class DataGenerator:
         self.to_idx = config["runner"]["generate"]["to"]
         self.set_status_path = f"http://localhost:{self.port}/project/set_status"
         self.log_path = f"http://localhost:{self.port}/project/add_log"
-        self.obj_name_list = os.listdir(self.obj_dir)[self.from_idx: self.to_idx]
+        self.origin_obj_name_list = os.listdir(self.obj_dir)[self.from_idx: self.to_idx]
+        if not os.path.exists(self.output_dir):
+            os.makedirs(self.output_dir)
+        self.obj_name_list = []
+        if self.use_list:
+            self.target_name_list = [line.strip() for line in open(self.object_list_path).readlines()]
+            for obj_name in self.target_name_list:
+                if obj_name in self.origin_obj_name_list:
+                    self.obj_name_list.append(obj_name)
+        else:
+            self.obj_name_list = self.origin_obj_name_list
+
         self.target_obj = None
         self.stopped = False
         self.random_obj_list = []
@@ -101,31 +116,8 @@ class DataGenerator:
         
         bpy.ops.rigidbody.object_add()
         bpy.context.object.rigid_body.type = 'PASSIVE'
-        bpy.ops.object.shade_auto_smooth()
         
-        # 创建不受光照影响的材质
-        mat = bpy.data.materials.new(name="RedMaterial")
-        mat.use_nodes = True
-        
-        # 清除默认节点
-        nodes = mat.node_tree.nodes
-        for node in nodes:
-            nodes.remove(node)
-
-        # 添加 Emission 节点
-        emission_node = nodes.new(type='ShaderNodeEmission')
-        emission_node.inputs['Color'].default_value = (1.0, 0.0, 0.0, 1.0)  # 红色
-
-        # 添加 Material Output 节点
-        output_node = nodes.new(type='ShaderNodeOutputMaterial')
-
-        # 连接节点
-        links = mat.node_tree.links
-        links.new(emission_node.outputs['Emission'], output_node.inputs['Surface'])
-
-        # 将材质赋给对象
-        platform.data.materials.clear()
-        platform.data.materials.append(mat)
+        MaterialUtil.change_object_material(platform, MaterialUtil.create_mask_material(color=(1.0, 0, 0)))
 
         self.display_table_config = {
             "height": height,
@@ -135,6 +127,7 @@ class DataGenerator:
         return platform
 
     def put_display_object(self, name):
+        
         config = self.random_config["display_object"]
 
         x = random.uniform(config["min_x"], config["max_x"])
@@ -154,10 +147,8 @@ class DataGenerator:
         platform_bbox = self.platform.bound_box
         platform_bbox_world = [self.platform.matrix_world @ mathutils.Vector(corner) for corner in platform_bbox]
         platform_top_z = max([v.z for v in platform_bbox_world])
-        
         obj_mesh_path = BlenderUtils.get_obj_path(self.obj_dir, name)
         obj = BlenderUtils.load_obj(name, obj_mesh_path)
-        
         obj_bottom_z = BlenderUtils.get_object_bottom_z(obj)
         offset_z = obj_bottom_z
         
@@ -166,31 +157,7 @@ class DataGenerator:
         
         bpy.ops.rigidbody.object_add()
         bpy.context.object.rigid_body.type = 'ACTIVE'
-
-        # 创建不受光照影响的材质
-        mat = bpy.data.materials.new(name="GreenMaterial")
-        mat.use_nodes = True
-        
-        # 清除默认节点
-        nodes = mat.node_tree.nodes
-        for node in nodes:
-            nodes.remove(node)
-
-        # 添加 Emission 节点
-        emission_node = nodes.new(type='ShaderNodeEmission')
-        emission_node.inputs['Color'].default_value = (0.0, 1.0, 0.0, 1.0)  # 绿色
-
-        # 添加 Material Output 节点
-        output_node = nodes.new(type='ShaderNodeOutputMaterial')
-
-        # 连接节点
-        links = mat.node_tree.links
-        links.new(emission_node.outputs['Emission'], output_node.inputs['Surface'])
-
-        # 将材质赋给对象
-        obj.data.materials.clear()
-        obj.data.materials.append(mat)
-        
+        MaterialUtil.change_object_material(obj, MaterialUtil.create_mask_material(color=(0, 1.0, 0)))
         self.target_obj = obj
 
     
@@ -243,11 +210,21 @@ class DataGenerator:
         np.savetxt(os.path.join(scene_dir, "points_and_normals.txt"), points_normals)
         for i, cam_pose in enumerate(view_data["cam_poses"]):
             BlenderUtils.set_camera_at(cam_pose)
-            BlenderUtils.render_and_save(scene_dir, f"{i}", binocular_vision=self.binocular_vision, target_object = self.target_obj)
+            BlenderUtils.render_mask(scene_dir, f"{i}", binocular_vision=self.binocular_vision, target_object = self.target_obj)
             BlenderUtils.save_cam_params(scene_dir, i, binocular_vision=self.binocular_vision)
             self.set_progress("render frame", i, len(view_data["cam_poses"]))
         self.set_progress("render frame", len(view_data["cam_poses"]), len(view_data["cam_poses"]))
         BlenderUtils.save_scene_info(scene_dir, self.display_table_config, object_name)
+        
+        MaterialUtil.change_object_material(self.target_obj, MaterialUtil.create_normal_material())
+        
+        for i, cam_pose in enumerate(view_data["cam_poses"]):
+            BlenderUtils.set_camera_at(cam_pose)
+            BlenderUtils.render_normal_and_depth(scene_dir, f"{i}", binocular_vision=self.binocular_vision, target_object = self.target_obj)
+            BlenderUtils.save_cam_params(scene_dir, i, binocular_vision=self.binocular_vision)
+            self.set_progress("render normal frame", i, len(view_data["cam_poses"]))
+        self.set_progress("render normal frame", len(view_data["cam_poses"]), len(view_data["cam_poses"]))
+        
         depth_dir = os.path.join(scene_dir, "depth")
         for depth_file in os.listdir(depth_dir):
             if not depth_file.endswith(".png"):
@@ -255,15 +232,10 @@ class DataGenerator:
                 file_path = os.path.join(depth_dir, depth_file)
                 new_file_path = os.path.join(depth_dir, f"{name}.png")
                 os.rename(file_path,new_file_path)
-        normal_dir = os.path.join(scene_dir, "normal")
-        for normal_file in os.listdir(normal_dir):
-            if not normal_file.endswith(".png"):
-                name, _ = os.path.splitext(normal_file)
-                file_path = os.path.join(normal_dir, normal_file)
-                new_file_path = os.path.join(normal_dir, f"{name}.png")
-                os.rename(file_path,new_file_path)
+        
         return True
     
+        
     def simulate_scene(self, frame_limit=120, depth = 0, diag = 0):
         bpy.context.view_layer.update()
         bpy.ops.screen.animation_play()
@@ -285,11 +257,7 @@ class DataGenerator:
 
         msg = self.check_and_adjust_target()
         
-        if msg == "adjusted" and depth < 3:
-            bpy.context.view_layer.update()
-            bpy.context.scene.frame_set(0)
-            return self.simulate_scene(depth = depth + 1, diag=diag)
-        elif msg == "success":
+        if msg == "success":
             print("Scene generation completed.")
             result = self.start_render(diag=diag)
             if not result:
@@ -307,11 +275,12 @@ class DataGenerator:
         if diag > self.max_diag or diag < self.min_diag:
             self.add_log(f"The diagonal size of the object <{object_name}>(size: {round(diag,3)}) does not meet the requirements.", "error")
             return "diag_error"
+        
         return self.simulate_scene(diag=diag)
 
     
     def gen_all_scene_data(self):
-        max_retry_times = 3
+        max_retry_times = 5
         total = len(self.obj_name_list)
         count = 0
         count_success = 0
@@ -324,11 +293,12 @@ class DataGenerator:
             if os.path.exists(scene_info_path):
                 self.add_log(f"Scene for object <{target_obj_name}> already exists, skipping", "warning")
                 count += 1
+                count_success += 1  
                 continue
             
             retry_times = 0
             self.set_status("target_object", target_obj_name)
-            while retry_times < 3 and result == "retry":
+            while retry_times < max_retry_times and result == "retry":
                 self.reset()  
                 try:
                     result = self.gen_scene_data(target_obj_name)

data_generator_2.py

@@ -0,0 +1,86 @@
+
+import os
+import bpy
+import numpy as np
+from utils.blender_util import BlenderUtils
+from utils.cad_view_sample_util import CADViewSampleUtil
+from utils.material_util import MaterialUtil
+
+class DataGenerator:
+    def __init__(self, config):
+        self.plane_size = config["runner"]["generate"]["plane_size"] 
+        self.output_dir = config["runner"]["generate"]["output_dir"]
+        self.random_config = config["runner"]["generate"]["random_config"]
+        self.light_and_camera_config = config["runner"]["generate"]["light_and_camera_config"]
+        self.max_views = config["runner"]["generate"]["max_views"]
+        self.min_views = config["runner"]["generate"]["min_views"]
+        self.min_diag = config["runner"]["generate"]["min_diag"]
+        self.max_diag = config["runner"]["generate"]["max_diag"]
+        self.binocular_vision = config["runner"]["generate"]["binocular_vision"]
+        self.target_obj = None
+        self.stopped = False
+        self.random_obj_list = []
+        self.display_table_config = {}
+        BlenderUtils.setup_scene(self.light_and_camera_config, None, self.binocular_vision)
+        self.table = BlenderUtils.get_obj(BlenderUtils.TABLE_NAME)
+        
+    def put_display_object(self, name):
+        obj_mesh_path = BlenderUtils.get_obj_path(self.obj_dir, name)
+        obj = BlenderUtils.load_obj(name, obj_mesh_path)
+        bpy.ops.rigidbody.object_add()
+        bpy.context.object.rigid_body.type = 'ACTIVE'
+        MaterialUtil.change_object_material(obj, MaterialUtil.create_mask_material(color=(0, 1.0, 0)))
+        self.target_obj = obj
+
+    def reset(self):
+        self.target_obj = None
+        self.random_obj_list = []
+        BlenderUtils.reset_objects_and_platform()
+    
+    def start_render(self, diag=0):
+        object_name = self.target_obj.name
+        if "." in object_name:
+            object_name = object_name.split(".")[0]
+        scene_dir = os.path.join(self.output_dir, object_name)
+        if not os.path.exists(scene_dir):
+            os.makedirs(scene_dir)
+        view_num = int(self.min_views + (diag - self.min_diag)/(self.max_diag - self.min_diag) * (self.max_views - self.min_views))
+        view_data = CADViewSampleUtil.sample_view_data_world_space(self.target_obj, distance_range=(0.25,0.5), voxel_size=0.005, max_views=view_num, min_cam_table_included_degree = self.min_cam_table_included_degree, random_view_ratio = self.random_view_ratio )
+        object_points = np.array(view_data["voxel_down_sampled_points"])
+        normals = np.array(view_data["normals"])
+        points_normals = np.concatenate((object_points, normals), axis=1)
+        
+        np.savetxt(os.path.join(scene_dir, "points_and_normals.txt"), points_normals)
+        for i, cam_pose in enumerate(view_data["cam_poses"]):
+            BlenderUtils.set_camera_at(cam_pose)
+            BlenderUtils.render_mask(scene_dir, f"{i}", binocular_vision=self.binocular_vision, target_object = self.target_obj)
+            BlenderUtils.save_cam_params(scene_dir, i, binocular_vision=self.binocular_vision)
+        BlenderUtils.save_scene_info(scene_dir, self.display_table_config, object_name)
+        
+        MaterialUtil.change_object_material(self.target_obj, MaterialUtil.create_normal_material())
+        
+        for i, cam_pose in enumerate(view_data["cam_poses"]):
+            BlenderUtils.set_camera_at(cam_pose)
+            BlenderUtils.render_normal_and_depth(scene_dir, f"{i}", binocular_vision=self.binocular_vision, target_object = self.target_obj)
+            BlenderUtils.save_cam_params(scene_dir, i, binocular_vision=self.binocular_vision)
+        
+        depth_dir = os.path.join(scene_dir, "depth")
+        for depth_file in os.listdir(depth_dir):
+            if not depth_file.endswith(".png"):
+                name, _ = os.path.splitext(depth_file)
+                file_path = os.path.join(depth_dir, depth_file)
+                new_file_path = os.path.join(depth_dir, f"{name}.png")
+                os.rename(file_path,new_file_path)
+        BlenderUtils.save_blend(scene_dir)
+        exit(0)
+        
+        return True
+
+    def gen_scene_data(self, object_name):
+        
+        bpy.context.scene.frame_set(0)
+        self.put_display_object(object_name)
+        diag = BlenderUtils.get_obj_diag(self.target_obj.name)
+        self.start_render(diag)
+    
+        

data_load.py

@@ -1,265 +0,0 @@
-import os
-import numpy as np
-import json
-import cv2
-import trimesh
-from pts import PtsUtil
-
-class DataLoadUtil:
-    
-    @staticmethod
-    def get_path(root, scene_name, frame_idx):
-        path = os.path.join(root, scene_name, f"{frame_idx}")
-        return path
-    
-    @staticmethod
-    def get_label_path(root, scene_name):
-        path = os.path.join(root,scene_name, f"label.json")
-        return path
-    
-    @staticmethod
-    def get_sampled_model_points_path(root, scene_name):
-        path = os.path.join(root,scene_name, f"sampled_model_points.txt")
-        return path
-    
-    @staticmethod
-    def get_scene_seq_length(root, scene_name):
-        camera_params_path = os.path.join(root, scene_name, "camera_params")
-        return len(os.listdir(camera_params_path))
-    
-    @staticmethod
-    def load_downsampled_world_model_points(root, scene_name):
-        model_path = DataLoadUtil.get_sampled_model_points_path(root, scene_name)
-        model_points = np.loadtxt(model_path)
-        return model_points
-
-    @staticmethod
-    def save_downsampled_world_model_points(root, scene_name, model_points):
-        model_path = DataLoadUtil.get_sampled_model_points_path(root, scene_name)
-        np.savetxt(model_path, model_points)
-    
-    @staticmethod
-    def load_mesh_at(model_dir, object_name, world_object_pose):
-        model_path = os.path.join(model_dir, object_name, "mesh.obj")
-        mesh = trimesh.load(model_path)
-        mesh.apply_transform(world_object_pose)
-        return mesh
-    
-    @staticmethod
-    def get_bbox_diag(model_dir, object_name):
-        model_path = os.path.join(model_dir, object_name, "mesh.obj")
-        mesh = trimesh.load(model_path)
-        bbox = mesh.bounding_box.extents
-        diagonal_length = np.linalg.norm(bbox)
-        return diagonal_length
-    
-    
-    @staticmethod
-    def save_mesh_at(model_dir, output_dir, object_name, scene_name, world_object_pose):
-        mesh = DataLoadUtil.load_mesh_at(model_dir, object_name, world_object_pose)
-        model_path = os.path.join(output_dir, scene_name, "world_mesh.obj")
-        mesh.export(model_path)
-
-    @staticmethod
-    def save_target_mesh_at_world_space(root, model_dir, scene_name):
-        scene_info = DataLoadUtil.load_scene_info(root, scene_name)
-        target_name = scene_info["target_name"]
-        transformation = scene_info[target_name]
-        location = transformation["location"]
-        rotation_euler = transformation["rotation_euler"]
-        pose_mat = trimesh.transformations.euler_matrix(*rotation_euler)
-        pose_mat[:3, 3] = location
-
-        mesh = DataLoadUtil.load_mesh_at(model_dir, target_name, pose_mat)
-        mesh_dir = os.path.join(root, scene_name, "mesh")
-        if not os.path.exists(mesh_dir):
-            os.makedirs(mesh_dir)
-        model_path = os.path.join(mesh_dir, "world_target_mesh.obj")
-        mesh.export(model_path)
-    
-    @staticmethod
-    def load_scene_info(root, scene_name):
-        scene_info_path = os.path.join(root, scene_name, "scene_info.json")
-        with open(scene_info_path, "r") as f:
-            scene_info = json.load(f)
-        return scene_info
-    
-    @staticmethod
-    def load_target_object_pose(root, scene_name):
-        scene_info = DataLoadUtil.load_scene_info(root, scene_name)
-        target_name = scene_info["target_name"]
-        transformation = scene_info[target_name]
-        location = transformation["location"]
-        rotation_euler = transformation["rotation_euler"]
-        pose_mat = trimesh.transformations.euler_matrix(*rotation_euler)
-        pose_mat[:3, 3] = location
-        return pose_mat
-    
-    @staticmethod
-    def load_depth(path, min_depth=0.01,max_depth=5.0,binocular=False):
-        
-        def load_depth_from_real_path(real_path, min_depth, max_depth):
-            depth = cv2.imread(real_path, cv2.IMREAD_UNCHANGED)
-            depth = depth.astype(np.float32) / 65535.0
-            min_depth = min_depth
-            max_depth = max_depth
-            depth_meters = min_depth + (max_depth - min_depth) * depth
-            return depth_meters
-        
-        if binocular:
-            depth_path_L = os.path.join(os.path.dirname(path), "depth", os.path.basename(path) + "_L.png")
-            depth_path_R = os.path.join(os.path.dirname(path), "depth", os.path.basename(path) + "_R.png")
-            depth_meters_L = load_depth_from_real_path(depth_path_L, min_depth, max_depth)
-            depth_meters_R = load_depth_from_real_path(depth_path_R, min_depth, max_depth)
-            return depth_meters_L, depth_meters_R
-        else:
-            depth_path = os.path.join(os.path.dirname(path), "depth", os.path.basename(path) + ".png")
-            depth_meters = load_depth_from_real_path(depth_path, min_depth, max_depth)
-            return depth_meters
-        
-    @staticmethod
-    def load_seg(path, binocular=False):
-        if binocular:
-            def clean_mask(mask_image):
-                green = [0, 255, 0, 255]
-                red = [255, 0, 0, 255]
-                threshold = 2
-                mask_image = np.where(np.abs(mask_image - green) <= threshold, green, mask_image)
-                mask_image = np.where(np.abs(mask_image - red) <= threshold, red, mask_image)
-                return mask_image
-            mask_path_L = os.path.join(os.path.dirname(path), "mask", os.path.basename(path) + "_L.png")
-            mask_image_L = clean_mask(cv2.imread(mask_path_L, cv2.IMREAD_UNCHANGED))
-            mask_path_R = os.path.join(os.path.dirname(path), "mask", os.path.basename(path) + "_R.png")
-            mask_image_R = clean_mask(cv2.imread(mask_path_R, cv2.IMREAD_UNCHANGED))
-            return mask_image_L, mask_image_R
-        else:
-            mask_path = os.path.join(os.path.dirname(path), "mask", os.path.basename(path) + ".png")
-            mask_image = cv2.imread(mask_path, cv2.IMREAD_GRAYSCALE)
-            return mask_image
-    
-    @staticmethod
-    def load_label(path):
-        with open(path, 'r') as f:
-            label_data = json.load(f)
-        return label_data
-    
-    @staticmethod
-    def load_rgb(path):
-        rgb_path = os.path.join(os.path.dirname(path), "rgb", os.path.basename(path) + ".png")
-        rgb_image = cv2.imread(rgb_path, cv2.IMREAD_COLOR)
-        return rgb_image
-    
-    @staticmethod
-    def cam_pose_transformation(cam_pose_before):
-        offset = np.asarray([
-            [1, 0, 0, 0],
-            [0, -1, 0, 0],
-            [0, 0, -1, 0],
-            [0, 0, 0, 1]])  
-        cam_pose_after =  cam_pose_before @ offset
-        return cam_pose_after
-    
-    @staticmethod
-    def load_cam_info(path, binocular=False):
-        camera_params_path = os.path.join(os.path.dirname(path), "camera_params", os.path.basename(path) + ".json")
-        with open(camera_params_path, 'r') as f:
-            label_data = json.load(f)
-        cam_to_world = np.asarray(label_data["extrinsic"])
-        cam_to_world = DataLoadUtil.cam_pose_transformation(cam_to_world)
-        cam_intrinsic = np.asarray(label_data["intrinsic"])
-        cam_info = {
-            "cam_to_world": cam_to_world,
-            "cam_intrinsic": cam_intrinsic,
-            "far_plane": label_data["far_plane"],
-            "near_plane": label_data["near_plane"]
-        }
-        if binocular:
-            cam_to_world_R = np.asarray(label_data["extrinsic_R"])
-            cam_to_world_R = DataLoadUtil.cam_pose_transformation(cam_to_world_R)
-            cam_info["cam_to_world_R"] = cam_to_world_R
-        return cam_info
-    
-    @staticmethod
-    def get_target_point_cloud(depth, cam_intrinsic, cam_extrinsic, mask, target_mask_label=(0,255,0,255)):
-        h, w = depth.shape
-        i, j = np.meshgrid(np.arange(w), np.arange(h), indexing='xy')
-        
-        z = depth
-        x = (i - cam_intrinsic[0, 2]) * z / cam_intrinsic[0, 0]
-        y = (j - cam_intrinsic[1, 2]) * z / cam_intrinsic[1, 1]
-        
-        points_camera = np.stack((x, y, z), axis=-1).reshape(-1, 3)
-        mask = mask.reshape(-1,4)
-
-        target_mask = (mask == target_mask_label).all(axis=-1)  
-        
-        target_points_camera = points_camera[target_mask]
-        target_points_camera_aug = np.concatenate([target_points_camera, np.ones((target_points_camera.shape[0], 1))], axis=-1)
-        
-        target_points_world = np.dot(cam_extrinsic, target_points_camera_aug.T).T[:, :3]
-        return {
-            "points_world": target_points_world,
-            "points_camera": target_points_camera
-        }
-    
-    @staticmethod
-    def get_point_cloud(depth, cam_intrinsic, cam_extrinsic):
-        h, w = depth.shape
-        i, j = np.meshgrid(np.arange(w), np.arange(h), indexing='xy')
-        
-        z = depth
-        x = (i - cam_intrinsic[0, 2]) * z / cam_intrinsic[0, 0]
-        y = (j - cam_intrinsic[1, 2]) * z / cam_intrinsic[1, 1]
-        
-        points_camera = np.stack((x, y, z), axis=-1).reshape(-1, 3)
-        points_camera_aug = np.concatenate([points_camera, np.ones((points_camera.shape[0], 1))], axis=-1)
-        
-        points_world = np.dot(cam_extrinsic, points_camera_aug.T).T[:, :3]
-        return {
-            "points_world": points_world,
-            "points_camera": points_camera
-        }
-        
-    @staticmethod
-    def get_target_point_cloud_world_from_path(path, binocular=False, random_downsample_N=65536, voxel_size = 0.005, target_mask_label=(0,255,0,255)):
-        cam_info = DataLoadUtil.load_cam_info(path, binocular=binocular)
-        if binocular:
-            depth_L, depth_R = DataLoadUtil.load_depth(path, cam_info['near_plane'], cam_info['far_plane'], binocular=True)
-            mask_L, mask_R = DataLoadUtil.load_seg(path, binocular=True)
-            point_cloud_L = DataLoadUtil.get_target_point_cloud(depth_L, cam_info['cam_intrinsic'], cam_info['cam_to_world'], mask_L, target_mask_label)['points_world']
-            point_cloud_R = DataLoadUtil.get_target_point_cloud(depth_R, cam_info['cam_intrinsic'], cam_info['cam_to_world_R'], mask_R, target_mask_label)['points_world']
-            point_cloud_L = PtsUtil.random_downsample_point_cloud(point_cloud_L, random_downsample_N)
-            point_cloud_R = PtsUtil.random_downsample_point_cloud(point_cloud_R, random_downsample_N)
-            overlap_points = DataLoadUtil.get_overlapping_points(point_cloud_L, point_cloud_R, voxel_size)
-            return overlap_points
-        else:
-            depth = DataLoadUtil.load_depth(path, cam_info['near_plane'], cam_info['far_plane'])
-            mask = DataLoadUtil.load_seg(path)
-            point_cloud = DataLoadUtil.get_target_point_cloud(depth, cam_info['cam_intrinsic'], cam_info['cam_to_world'], mask)['points_world']
-            return point_cloud
-            
-    
-    @staticmethod
-    def voxelize_points(points, voxel_size):
- 
-        voxel_indices = np.floor(points / voxel_size).astype(np.int32)
-        unique_voxels = np.unique(voxel_indices, axis=0, return_inverse=True)
-        return unique_voxels
-    
-    @staticmethod
-    def get_overlapping_points(point_cloud_L, point_cloud_R, voxel_size=0.005):
-        voxels_L, indices_L = DataLoadUtil.voxelize_points(point_cloud_L, voxel_size)
-        voxels_R, _ = DataLoadUtil.voxelize_points(point_cloud_R, voxel_size)
-
-        voxel_indices_L = voxels_L.view([('', voxels_L.dtype)]*3)
-        voxel_indices_R = voxels_R.view([('', voxels_R.dtype)]*3)
-        overlapping_voxels = np.intersect1d(voxel_indices_L, voxel_indices_R)
-        mask_L = np.isin(indices_L, np.where(np.isin(voxel_indices_L, overlapping_voxels))[0])
-        overlapping_points = point_cloud_L[mask_L]
-        return overlapping_points
-    
-    @staticmethod
-    def load_points_normals(root, scene_name):
-        points_path = os.path.join(root, scene_name, "points_and_normals.txt")
-        points_normals = np.loadtxt(points_path)
-        return points_normals

data_renderer.py

@@ -2,11 +2,12 @@ import os
 import bpy
 import sys
 import json
+import time
 import mathutils
 import numpy as np
 sys.path.append(os.path.dirname(os.path.abspath(__file__)))
-from blender_util import BlenderUtils
-
+from utils.blender_util import BlenderUtils
+from utils.material_util import MaterialUtil
 
 class DataRenderer:
     def __init__(self):
@@ -23,8 +24,8 @@ class DataRenderer:
                                 "near_plane": 0.01,
                                 "far_plane": 5,
                                 "fov_vertical": 25,
-                                "resolution": [1280, 800],
-                                "eye_distance": 0.15,
+                                "resolution": [640, 400],
+                                "eye_distance": 0.10,
                                 "eye_angle": 25
                             },
                             "Light": {
@@ -42,7 +43,6 @@ class DataRenderer:
         self.light_and_camera_config = config["renderer"]["generate"]["light_and_camera_config"]
         self.obj_dir = config["renderer"]["generate"]["object_dir"]
         self.binocular_vision = config["renderer"]["generate"]["binocular_vision"]
-        self.obj_name_list = os.listdir(self.obj_dir)
         self.target_obj = None
         self.random_obj_list = []
 
@@ -58,14 +58,31 @@ class DataRenderer:
     def do_render(self, cam_pose, restore_info, temp_dir):
         
         self.reset()
+        start_time = time.time()
         self.restore_scene(restore_info=restore_info)
+        end_time = time.time()
+        print(f"Time taken for restoring scene: {end_time - start_time} seconds")
         object_name = self.target_obj.name
         temp_file_name = f"tmp"
         if "." in object_name:
             object_name = object_name.split(".")[0]
+        start_time = time.time()
         BlenderUtils.set_camera_at(cam_pose)
-        BlenderUtils.render_and_save(temp_dir, temp_file_name, binocular_vision=self.binocular_vision)
+        BlenderUtils.render_mask(temp_dir, temp_file_name, binocular_vision=self.binocular_vision)
+        MaterialUtil.change_object_material(self.target_obj, MaterialUtil.create_normal_material())
+        BlenderUtils.render_normal_and_depth(temp_dir, temp_file_name, binocular_vision=self.binocular_vision)
+        end_time = time.time()
+        print(f"Time taken for rendering: {end_time - start_time} seconds")
         
+        BlenderUtils.save_cam_params(temp_dir, temp_file_name, binocular_vision=self.binocular_vision)
+        
+        depth_dir = os.path.join(temp_dir, "depth")
+        for depth_file in os.listdir(depth_dir):
+            if not depth_file.endswith(".png"):
+                name, _ = os.path.splitext(depth_file)
+                file_path = os.path.join(depth_dir, depth_file)
+                new_file_path = os.path.join(depth_dir, f"{name}.png")
+                os.rename(file_path,new_file_path)
                 
     def restore_scene(self, restore_info):
 
@@ -82,12 +99,7 @@ class DataRenderer:
                     obj.location = mathutils.Vector(obj_info["location"])
                     obj.rotation_euler = mathutils.Vector(obj_info["rotation_euler"])
                     obj.scale = mathutils.Vector(obj_info["scale"])
-                    mat = bpy.data.materials.new(name="GreenMaterial")
-                    mat.diffuse_color = (0.0, 1.0, 0.0, 1.0)  # Green with full alpha (1.0)
-                    if len(obj.data.materials) > 0:
-                        obj.data.materials[0] = mat
-                    else:
-                        obj.data.materials.append(mat)
+                    MaterialUtil.change_object_material(obj, MaterialUtil.create_mask_material(color=(0, 1.0, 0)))
                     self.target_obj = obj
 
     def restore_display_platform(self, platform_info):
@@ -96,16 +108,9 @@ class DataRenderer:
         platform.name = BlenderUtils.DISPLAY_TABLE_NAME
         platform.location = mathutils.Vector(platform_info["location"])
 
-        mat = bpy.data.materials.new(name="RedMaterial")
-        mat.diffuse_color = (1.0, 0.0, 0.0, 1.0)  # Red with full alpha (1.0)
-        if len(platform.data.materials) > 0:
-            platform.data.materials[0] = mat
-        else:
-            platform.data.materials.append(mat)
-
         bpy.ops.rigidbody.object_add()
         bpy.context.object.rigid_body.type = 'PASSIVE'
-        bpy.ops.object.shade_auto_smooth()
+        MaterialUtil.change_object_material(platform, MaterialUtil.create_mask_material(color=(1.0, 0, 0)))
 
 def main(temp_dir):
     params_data_path = os.path.join(temp_dir, "params.json")
@@ -116,9 +121,14 @@ def main(temp_dir):
     scene_info_path = os.path.join(params_data["scene_path"], "scene_info.json")
     with open(scene_info_path, 'r') as f:
         scene_info = json.load(f)
-    
+    start_time = time.time()
     data_renderer = DataRenderer()
+    end_time = time.time()
+    print(f"Time taken for initialization: {end_time - start_time} seconds")
+    start_time = time.time()
     data_renderer.do_render(cam_pose, scene_info, temp_dir)
+    end_time = time.time()
+    print(f"Time taken for rendering: {end_time - start_time} seconds")
     depth_dir = os.path.join(temp_dir, "depth")
     for depth_file in os.listdir(depth_dir):
         if not depth_file.endswith(".png"):

pts.py

@@ -1,22 +0,0 @@
-import numpy as np
-import open3d as o3d
-
-class PtsUtil:
-
-    @staticmethod
-    def voxel_downsample_point_cloud(point_cloud, voxel_size=0.005):
-        o3d_pc = o3d.geometry.PointCloud()
-        o3d_pc.points = o3d.utility.Vector3dVector(point_cloud)
-        downsampled_pc = o3d_pc.voxel_down_sample(voxel_size)
-        return np.asarray(downsampled_pc.points)
-    
-    @staticmethod
-    def transform_point_cloud(points, pose_mat):
-        points_h = np.concatenate([points, np.ones((points.shape[0], 1))], axis=1)
-        points_h = np.dot(pose_mat, points_h.T).T
-        return points_h[:, :3]
-    
-    @staticmethod
-    def random_downsample_point_cloud(point_cloud, num_points):
-        idx = np.random.choice(len(point_cloud), num_points, replace=True)
-        return point_cloud[idx]

reconstruction.py

@@ -1,119 +0,0 @@
-import numpy as np
-from scipy.spatial import cKDTree
-from pts import PtsUtil
-
-class ReconstructionUtil:
-    
-    @staticmethod
-    def compute_coverage_rate(target_point_cloud, combined_point_cloud, threshold=0.01):
-        kdtree = cKDTree(combined_point_cloud)
-        distances, _ = kdtree.query(target_point_cloud)
-        covered_points = np.sum(distances < threshold)
-        coverage_rate = covered_points / target_point_cloud.shape[0]
-        return coverage_rate
-    
-    @staticmethod
-    def compute_overlap_rate(new_point_cloud, combined_point_cloud, threshold=0.01):
-        kdtree = cKDTree(combined_point_cloud)
-        distances, _ = kdtree.query(new_point_cloud)
-        overlapping_points = np.sum(distances < threshold)
-        overlap_rate = overlapping_points / new_point_cloud.shape[0]
-        return overlap_rate
-    
-    @staticmethod
-    def combine_point_with_view_sequence(point_list, view_sequence):
-        selected_views = []
-        for view_index, _ in view_sequence:
-            selected_views.append(point_list[view_index])
-        return np.vstack(selected_views)
-    
-    @staticmethod 
-    def compute_next_view_coverage_list(views, combined_point_cloud, target_point_cloud, threshold=0.01):
-        best_view = None
-        best_coverage_increase = -1
-        current_coverage = ReconstructionUtil.compute_coverage_rate(target_point_cloud, combined_point_cloud, threshold)
-        
-        for view_index, view in enumerate(views):
-            candidate_views = combined_point_cloud + [view]
-            down_sampled_combined_point_cloud = PtsUtil.voxel_downsample_point_cloud(candidate_views, threshold)
-            new_coverage = ReconstructionUtil.compute_coverage_rate(target_point_cloud, down_sampled_combined_point_cloud, threshold)
-            coverage_increase = new_coverage - current_coverage
-            if coverage_increase > best_coverage_increase:
-                best_coverage_increase = coverage_increase
-                best_view = view_index
-        return best_view, best_coverage_increase
-
-    
-    @staticmethod
-    def compute_next_best_view_sequence_with_overlap(target_point_cloud, point_cloud_list, display_table_point_cloud_list = None,threshold=0.01, overlap_threshold=0.3, status_info=None):
-        selected_views = []
-        current_coverage = 0.0
-        remaining_views = list(range(len(point_cloud_list)))
-        view_sequence = []
-        cnt_processed_view = 0
-        while remaining_views:
-            best_view = None
-            best_coverage_increase = -1
-        
-            for view_index in remaining_views:
-                
-                if selected_views:
-                    combined_old_point_cloud = np.vstack(selected_views)
-                    down_sampled_old_point_cloud = PtsUtil.voxel_downsample_point_cloud(combined_old_point_cloud,threshold)
-                    down_sampled_new_view_point_cloud = PtsUtil.voxel_downsample_point_cloud(point_cloud_list[view_index],threshold)
-                    overlap_rate = ReconstructionUtil.compute_overlap_rate(down_sampled_new_view_point_cloud,down_sampled_old_point_cloud, threshold)
-                    if overlap_rate < overlap_threshold:
-                        continue
-                    
-                candidate_views = selected_views + [point_cloud_list[view_index]]
-                combined_point_cloud = np.vstack(candidate_views)
-                down_sampled_combined_point_cloud = PtsUtil.voxel_downsample_point_cloud(combined_point_cloud,threshold)
-                new_coverage = ReconstructionUtil.compute_coverage_rate(target_point_cloud, down_sampled_combined_point_cloud, threshold)
-                coverage_increase = new_coverage - current_coverage
-                #print(f"view_index: {view_index}, coverage_increase: {coverage_increase}")
-                if coverage_increase > best_coverage_increase:
-                    best_coverage_increase = coverage_increase
-                    best_view = view_index
-                    
-            
-            if best_view is not None:
-                if best_coverage_increase <=1e-3:
-                    break
-                selected_views.append(point_cloud_list[best_view])
-                remaining_views.remove(best_view)
-                current_coverage += best_coverage_increase
-                cnt_processed_view += 1
-                if status_info is not None:
-                    sm = status_info["status_manager"]
-                    app_name = status_info["app_name"]
-                    runner_name = status_info["runner_name"]
-                    sm.set_status(app_name, runner_name, "current coverage", current_coverage)
-                    sm.set_progress(app_name, runner_name, "processed view", cnt_processed_view, len(point_cloud_list))
-            
-                view_sequence.append((best_view, current_coverage))
-                
-            else:
-                break
-        if status_info is not None:
-            sm = status_info["status_manager"]
-            app_name = status_info["app_name"]
-            runner_name = status_info["runner_name"]
-            sm.set_progress(app_name, runner_name, "processed view", len(point_cloud_list), len(point_cloud_list))
-        return view_sequence, remaining_views, down_sampled_combined_point_cloud
-    
-    @staticmethod
-    def filter_points(points, points_normals, cam_pose,  voxel_size=0.005, theta=45):
-        sampled_points = PtsUtil.voxel_downsample_point_cloud(points, voxel_size)
-        kdtree = cKDTree(points_normals[:,:3])
-        _, indices = kdtree.query(sampled_points)
-        nearest_points = points_normals[indices]
-
-        normals = nearest_points[:, 3:]
-        camera_axis = -cam_pose[:3, 2] 
-        normals_normalized = normals / np.linalg.norm(normals, axis=1, keepdims=True)
-        cos_theta = np.dot(normals_normalized, camera_axis)
-        theta_rad = np.deg2rad(theta)
-        filtered_sampled_points= sampled_points[cos_theta > np.cos(theta_rad)]
-        
-        return filtered_sampled_points[:, :3]
-    

run_blender.py

@@ -1,10 +1,10 @@
 
 import os
 import sys
-sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
-sys.path.append("/home/hofee/.local/lib/python3.11/site-packages")
 import yaml
-from blender.data_generator import DataGenerator
+
+sys.path.append(os.path.dirname(os.path.abspath(__file__)))
+from data_generator import DataGenerator
 
 if __name__ == "__main__":
     config_path = sys.argv[sys.argv.index('--') + 1]

test/tempdir/params.json

@@ -1,29 +0,0 @@
-{
-    "cam_pose": [
-        [
-            -0.8127143979072571,
-            -0.3794165253639221,
-            0.4421972334384918,
-            0.2740877568721771
-        ],
-        [
-            0.5826622247695923,
-            -0.5292212963104248,
-            0.616789698600769,
-            0.46910107135772705
-        ],
-        [
-            0.0,
-            0.7589254975318909,
-            0.6511774659156799,
-            1.2532192468643188
-        ],
-        [
-            0.0,
-            0.0,
-            0.0,
-            1.0
-        ]
-    ],
-    "scene_path": "/media/hofee/data/project/python/nbv_reconstruction/sample_for_training/scenes/google_scan-backpack_0288"
-}

utils/blender_util.py

@@ -1,4 +1,3 @@
-
 import os
 import json
 import bpy
@@ -7,6 +6,7 @@ import gc
 import numpy as np
 import mathutils
 
+
 class BlenderUtils:
 
     TABLE_NAME: str = "table"
@@ -35,28 +35,27 @@ class BlenderUtils:
     def get_obj(name):
         return bpy.data.objects.get(name)
 
-            
     @staticmethod
     def get_obj_pose(name):
         obj = BlenderUtils.get_obj(name)
         return np.asarray(obj.matrix_world)
 
-    
-    
     @staticmethod
     def add_plane(name, location, orientation, size=10):
-        bpy.ops.mesh.primitive_plane_add(size=size,location=location)
+        bpy.ops.mesh.primitive_plane_add(size=size, location=location)
         plane = bpy.context.selected_objects[-1]
         plane.name = name
         plane.rotation_euler = orientation
         bpy.ops.rigidbody.object_add()
-        bpy.context.object.rigid_body.type = 'PASSIVE'
+        bpy.context.object.rigid_body.type = "PASSIVE"
 
     @staticmethod
     def add_table(table_model_path):
-        table = BlenderUtils.load_obj(BlenderUtils.TABLE_NAME, table_model_path, scale=0.01)
+        table = BlenderUtils.load_obj(
+            BlenderUtils.TABLE_NAME, table_model_path, scale=0.01
+        )
         bpy.ops.rigidbody.object_add()
-        bpy.context.object.rigid_body.type = 'PASSIVE'
+        bpy.context.object.rigid_body.type = "PASSIVE"
 
         mat = bpy.data.materials.new(name="TableYellowMaterial")
         mat.diffuse_color = (1.0, 1.0, 0.0, 1.0)
@@ -67,33 +66,42 @@ class BlenderUtils:
 
     @staticmethod
     def setup_scene(init_light_and_camera_config, table_model_path, binocular_vision):
-        bpy.context.scene.render.engine = 'BLENDER_EEVEE_NEXT'
+        bpy.context.scene.render.engine = "BLENDER_EEVEE"
         bpy.context.scene.display.shading.show_xray = False
         bpy.context.scene.display.shading.use_dof = False
-        bpy.context.scene.display.render_aa = 'OFF'
-        bpy.context.scene.view_settings.view_transform = 'Standard'
+        bpy.context.scene.display.render_aa = "OFF"
+        bpy.context.scene.view_settings.view_transform = "Standard"
 
-        bpy.context.scene.eevee.use_ssr = False            # 关闭屏幕空间反射
-        bpy.context.scene.eevee.use_bloom = False          # 关闭辉光
-        bpy.context.scene.eevee.use_gtao = False           # 关闭环境光遮蔽
-        bpy.context.scene.eevee.use_soft_shadows = False   # 关闭软阴影
+        bpy.context.scene.eevee.use_ssr = False  # 关闭屏幕空间反射
+        bpy.context.scene.eevee.use_bloom = False  # 关闭辉光
+        bpy.context.scene.eevee.use_gtao = False  # 关闭环境光遮蔽
+        bpy.context.scene.eevee.use_soft_shadows = False  # 关闭软阴影
         bpy.context.scene.eevee.use_shadows = False  # 关闭所有阴影
         bpy.context.scene.world.use_nodes = False  # 如果你不需要环境光，关闭环境节点
 
-
-        #bpy.context.scene.eevee.use_sss = False            # 关闭次表面散射
+        # bpy.context.scene.eevee.use_sss = False            # 关闭次表面散射
 
         # 2. 设置最低的采样数
         bpy.context.scene.eevee.taa_render_samples = 1
         bpy.context.scene.eevee.taa_samples = 1
-        BlenderUtils.init_light_and_camera(init_light_and_camera_config, binocular_vision)
+        BlenderUtils.init_light_and_camera(
+            init_light_and_camera_config, binocular_vision
+        )
 
-        BlenderUtils.add_plane("plane_floor", location=(0,0,0), orientation=(0,0,0))
-        BlenderUtils.add_plane("plane_ceil", location=(0,0,10), orientation=(0,0,0))
-        BlenderUtils.add_plane("plane_wall_1", location=(5,0,5), orientation=(0,np.pi/2,0))
-        BlenderUtils.add_plane("plane_wall_2", location=(-5,0,5), orientation=(0,np.pi/2,0))
-        BlenderUtils.add_plane("plane_wall_3", location=(0,5,5), orientation=(np.pi/2,0,0))
-        BlenderUtils.add_plane("plane_wall_4", location=(0,-5,5), orientation=(np.pi/2,0,0))
+        BlenderUtils.add_plane("plane_floor", location=(0, 0, 0), orientation=(0, 0, 0))
+        BlenderUtils.add_plane("plane_ceil", location=(0, 0, 10), orientation=(0, 0, 0))
+        BlenderUtils.add_plane(
+            "plane_wall_1", location=(5, 0, 5), orientation=(0, np.pi / 2, 0)
+        )
+        BlenderUtils.add_plane(
+            "plane_wall_2", location=(-5, 0, 5), orientation=(0, np.pi / 2, 0)
+        )
+        BlenderUtils.add_plane(
+            "plane_wall_3", location=(0, 5, 5), orientation=(np.pi / 2, 0, 0)
+        )
+        BlenderUtils.add_plane(
+            "plane_wall_4", location=(0, -5, 5), orientation=(np.pi / 2, 0, 0)
+        )
 
         BlenderUtils.add_table(table_model_path)
 
@@ -101,9 +109,9 @@ class BlenderUtils:
     def set_light_params(light, config):
         light.location = config["location"]
         light.rotation_euler = config["orientation"]
-        if light.type == 'SUN':
+        if light.type == "SUN":
             light.data.energy = config["power"]
-        elif light.type == 'POINT':
+        elif light.type == "POINT":
             light.data.energy = config["power"]
 
     @staticmethod
@@ -112,19 +120,19 @@ class BlenderUtils:
         camera_object = bpy.data.objects.new(BlenderUtils.CAMERA_OBJECT_NAME, None)
         bpy.context.collection.objects.link(camera_object)
         cameras = [bpy.data.objects.get("Camera")]
-        camera.location = [0,0,0]
-        camera.rotation_euler = [0,0,0]
+        camera.location = [0, 0, 0]
+        camera.rotation_euler = [0, 0, 0]
         camera.parent = camera_object
         if binocular_vision:
             left_camera = cameras[0]
             right_camera = left_camera.copy()
             right_camera.name = BlenderUtils.CAMERA_RIGHT_NAME
-            right_camera.data  = left_camera.data.copy()
+            right_camera.data = left_camera.data.copy()
             right_camera.data.name = BlenderUtils.CAMERA_RIGHT_NAME
             bpy.context.collection.objects.link(right_camera)
             right_camera.parent = camera_object
-            right_camera.location = [config["eye_distance"]/2, 0, 0]
-            left_camera.location = [-config["eye_distance"]/2, 0, 0]    
+            right_camera.location = [config["eye_distance"] / 2, 0, 0]
+            left_camera.location = [-config["eye_distance"] / 2, 0, 0]
             binocular_angle = config["eye_angle"]
             half_angle = np.radians(binocular_angle / 2)
 
@@ -139,16 +147,24 @@ class BlenderUtils:
             bpy.context.scene.render.resolution_x = config["resolution"][0]
             bpy.context.scene.render.resolution_y = config["resolution"][1]
             sensor_height = 24.0
-            focal_length = sensor_height / (2 * np.tan(np.radians(config["fov_vertical"]) / 2))
+            focal_length = sensor_height / (
+                2 * np.tan(np.radians(config["fov_vertical"]) / 2)
+            )
             camera.data.lens = focal_length
-            camera.data.sensor_width = sensor_height * config["resolution"][0] / config["resolution"][1]
+            camera.data.sensor_width = (
+                sensor_height * config["resolution"][0] / config["resolution"][1]
+            )
             camera.data.sensor_height = sensor_height
 
     @staticmethod
     def init_light_and_camera(init_light_and_camera_config, binocular_vision):
 
         camera = BlenderUtils.get_obj(BlenderUtils.CAMERA_NAME)
-        BlenderUtils.set_camera_params(camera, init_light_and_camera_config[BlenderUtils.CAMERA_NAME], binocular_vision)
+        BlenderUtils.set_camera_params(
+            camera,
+            init_light_and_camera_config[BlenderUtils.CAMERA_NAME],
+            binocular_vision,
+        )
 
     @staticmethod
     def get_obj_diag(name):
@@ -179,100 +195,107 @@ class BlenderUtils:
         return min_z
 
     @staticmethod
-    def render_and_save(output_dir, file_name, binocular_vision=False, target_object=None):
+    def render_normal_and_depth(
+        output_dir, file_name, binocular_vision=False, target_object=None
+    ):
+        # use pass z
+        bpy.context.scene.view_layers["ViewLayer"].use_pass_z = True
         target_cameras = [BlenderUtils.CAMERA_NAME]
         if binocular_vision:
             target_cameras.append(BlenderUtils.CAMERA_RIGHT_NAME)
+            
         for cam_name in target_cameras:
             bpy.context.scene.camera = BlenderUtils.get_obj(cam_name)
-            bpy.context.scene.view_layers["ViewLayer"].use_pass_z = True
-            bpy.context.scene.view_layers["ViewLayer"].use_pass_normal = True
             cam_suffix = "L" if cam_name == BlenderUtils.CAMERA_NAME else "R"
             scene = bpy.context.scene
             scene.render.filepath = ""
 
-            
-            mask_dir = os.path.join(output_dir, "mask")
+            mask_dir = os.path.join(output_dir, "normal")
             if not os.path.exists(mask_dir):
                 os.makedirs(mask_dir)
 
-            scene.render.filepath = os.path.join(output_dir, mask_dir, f"{file_name}_{cam_suffix}.png")
-            scene.render.image_settings.color_depth = '8'
+            scene.render.filepath = os.path.join(
+                output_dir, mask_dir, f"{file_name}_{cam_suffix}.exr"
+            ) 
+            
+            scene.render.image_settings.file_format = "OPEN_EXR"
+            scene.render.image_settings.color_mode = "RGB"
+            bpy.context.scene.view_settings.view_transform = "Raw"
+            scene.render.image_settings.color_depth = "16"
+            bpy.context.scene.render.filter_size = 1.5
             scene.render.resolution_percentage = 100
             scene.render.use_overwrite = False
             scene.render.use_file_extension = False
             scene.render.use_placeholder = False
-
             scene.use_nodes = True
             tree = scene.node_tree
 
             for node in tree.nodes:
                 tree.nodes.remove(node)
 
-            rl = tree.nodes.new('CompositorNodeRLayers')
+            rl = tree.nodes.new("CompositorNodeRLayers")
 
-            map_range = tree.nodes.new('CompositorNodeMapRange')
-            map_range.inputs['From Min'].default_value = 0.01
-            map_range.inputs['From Max'].default_value = 5
-            map_range.inputs['To Min'].default_value = 0 
-            map_range.inputs['To Max'].default_value = 1
-            tree.links.new(rl.outputs['Depth'], map_range.inputs[0])
+            map_range = tree.nodes.new("CompositorNodeMapRange")
+            map_range.inputs["From Min"].default_value = 0.01
+            map_range.inputs["From Max"].default_value = 5
+            map_range.inputs["To Min"].default_value = 0
+            map_range.inputs["To Max"].default_value = 1
+            tree.links.new(rl.outputs["Depth"], map_range.inputs[0])
 
-            output_depth = tree.nodes.new('CompositorNodeOutputFile')
+            output_depth = tree.nodes.new("CompositorNodeOutputFile")
 
             depth_dir = os.path.join(output_dir, "depth")
             if not os.path.exists(depth_dir):
                 os.makedirs(depth_dir)
             output_depth.base_path = depth_dir
             output_depth.file_slots[0].path = f"{file_name}_{cam_suffix}.####"
-            output_depth.format.file_format = 'PNG'
-            output_depth.format.color_mode = 'BW'
-            output_depth.format.color_depth = '16'
-
-           # 创建 Separate XYZ 节点来分离法线的 X, Y, Z 分量
-            separate_xyz = tree.nodes.new('CompositorNodeSeparateXYZ')
-
-            # 将法线向量连接到 Separate XYZ 节点
-            tree.links.new(rl.outputs['Normal'], separate_xyz.inputs[0])
-
-            # 创建 Map Range 节点来分别映射 X, Y, Z 分量
-            map_range_x = tree.nodes.new('CompositorNodeMapRange')
-            map_range_y = tree.nodes.new('CompositorNodeMapRange')
-            map_range_z = tree.nodes.new('CompositorNodeMapRange')
-
-            # 设置映射范围
-            for map_range in [map_range_x, map_range_y, map_range_z]:
-                map_range.inputs['From Min'].default_value = -1
-                map_range.inputs['From Max'].default_value = 1
-                map_range.inputs['To Min'].default_value = 0
-                map_range.inputs['To Max'].default_value = 1
-
-            # 分别连接到法线的 X, Y, Z 输出
-            tree.links.new(separate_xyz.outputs['X'], map_range_x.inputs[0])
-            tree.links.new(separate_xyz.outputs['Y'], map_range_y.inputs[0])
-            tree.links.new(separate_xyz.outputs['Z'], map_range_z.inputs[0])
-
-            # 合并 X, Y, Z 分量到一个 RGB 输出
-            combine_rgb = tree.nodes.new('CompositorNodeCombineXYZ')
-            tree.links.new(map_range_x.outputs[0], combine_rgb.inputs['X'])
-            tree.links.new(map_range_y.outputs[0], combine_rgb.inputs['Y'])
-            tree.links.new(map_range_z.outputs[0], combine_rgb.inputs['Z'])
-
-            # 输出到文件
-            output_normal = tree.nodes.new('CompositorNodeOutputFile')
-            normal_dir = os.path.join(output_dir, "normal")
-            if not os.path.exists(normal_dir):
-                os.makedirs(normal_dir)
-            output_normal.base_path = normal_dir
-            output_normal.file_slots[0].path = f"{file_name}_{cam_suffix}.####"
-            output_normal.format.file_format = 'PNG'
-            output_normal.format.color_mode = 'RGB'
-            output_normal.format.color_depth = '8'
-
-            tree.links.new(combine_rgb.outputs[0], output_normal.inputs[0])
+            output_depth.format.file_format = "PNG"
+            output_depth.format.color_mode = "BW"
+            output_depth.format.color_depth = "16"
+            tree.links.new(map_range.outputs[0], output_depth.inputs[0])
             bpy.ops.render.render(write_still=True)
 
+        msg = "success"
+        return msg
         
+    @staticmethod
+    def render_mask(
+        output_dir, file_name, binocular_vision=False, target_object=None
+    ):
+        target_cameras = [BlenderUtils.CAMERA_NAME]
+        if binocular_vision:
+            target_cameras.append(BlenderUtils.CAMERA_RIGHT_NAME)
+        
+        for cam_name in target_cameras:
+            bpy.context.scene.camera = BlenderUtils.get_obj(cam_name)
+            cam_suffix = "L" if cam_name == BlenderUtils.CAMERA_NAME else "R"
+            scene = bpy.context.scene
+            scene.render.filepath = ""
+
+            mask_dir = os.path.join(output_dir, "mask")
+            if not os.path.exists(mask_dir):
+                os.makedirs(mask_dir)
+
+            scene.render.filepath = os.path.join(
+                output_dir, mask_dir, f"{file_name}_{cam_suffix}.png"
+            )
+            scene.use_nodes = True
+            tree = scene.node_tree
+
+            for node in tree.nodes:
+                tree.nodes.remove(node)
+
+            rl = tree.nodes.new("CompositorNodeRLayers")
+            scene.render.image_settings.file_format = "PNG"
+            scene.render.image_settings.color_mode = "RGB"
+            scene.render.image_settings.color_depth = "8"
+            scene.render.resolution_percentage = 100
+            scene.render.use_overwrite = False
+            scene.render.use_file_extension = False
+            scene.render.use_placeholder = False
+
+            
+            bpy.ops.render.render(write_still=True)
 
         msg = "success"
         return msg
@@ -289,11 +312,11 @@ class BlenderUtils:
         resolution_y = bpy.context.scene.render.resolution_y
         intrinsic = np.zeros((3, 3))
         intrinsic[0, 0] = focal_length * resolution_x / sensor_width  # fx
-        intrinsic[1, 1] = focal_length * resolution_y / sensor_height   # fy
+        intrinsic[1, 1] = focal_length * resolution_y / sensor_height  # fy
         intrinsic[0, 2] = resolution_x / 2.0  # cx
         intrinsic[1, 2] = resolution_y / 2.0  # cy
         intrinsic[2, 2] = 1.0
-        cam_object =  BlenderUtils.get_obj(BlenderUtils.CAMERA_OBJECT_NAME)
+        cam_object = BlenderUtils.get_obj(BlenderUtils.CAMERA_OBJECT_NAME)
         extrinsic_cam_object = np.array(cam_object.matrix_world)
         data = {
             "extrinsic": extrinsic.tolist(),
@@ -305,21 +328,28 @@ class BlenderUtils:
         if binocular_vision:
             right_camera = BlenderUtils.get_obj(BlenderUtils.CAMERA_RIGHT_NAME)
             extrinsic_right = np.array(right_camera.matrix_world)
-            print("result:",extrinsic_right)
+            print("result:", extrinsic_right)
 
             data["extrinsic_R"] = extrinsic_right.tolist()
 
         cam_params_dir = os.path.join(scene_dir, "camera_params")
         if not os.path.exists(cam_params_dir):
             os.makedirs(cam_params_dir)
-        cam_params_path = os.path.join(cam_params_dir,  f"{idx}.json")
+        cam_params_path = os.path.join(cam_params_dir, f"{idx}.json")
         with open(cam_params_path, "w") as f:
             json.dump(data, f, indent=4)
 
     @staticmethod
     def reset_objects_and_platform():
         all_objects = bpy.data.objects
-        keep_objects = {"plane_floor", "plane_ceil", "plane_wall_1", "plane_wall_2", "plane_wall_3", "plane_wall_4"}
+        keep_objects = {
+            "plane_floor",
+            "plane_ceil",
+            "plane_wall_1",
+            "plane_wall_2",
+            "plane_wall_3",
+            "plane_wall_4",
+        }
         keep_objects.add(BlenderUtils.CAMERA_OBJECT_NAME)
         keep_objects.add(BlenderUtils.CAMERA_NAME)
         keep_objects.add(BlenderUtils.CAMERA_RIGHT_NAME)
@@ -345,18 +375,28 @@ class BlenderUtils:
     @staticmethod
     def save_scene_info(scene_root_dir, display_table_config, target_name):
         all_objects = bpy.data.objects
-        no_save_objects = {"plane_floor", "plane_ceil", "plane_wall_1", "plane_wall_2", "plane_wall_3", "plane_wall_4"}
+        no_save_objects = {
+            "plane_floor",
+            "plane_ceil",
+            "plane_wall_1",
+            "plane_wall_2",
+            "plane_wall_3",
+            "plane_wall_4",
+        }
         no_save_objects.add(BlenderUtils.CAMERA_OBJECT_NAME)
         no_save_objects.add(BlenderUtils.CAMERA_NAME)
         no_save_objects.add(BlenderUtils.CAMERA_RIGHT_NAME)
         no_save_objects.add(BlenderUtils.TABLE_NAME)
         scene_info = {}
         for obj in all_objects:
-            if obj.name not in no_save_objects and obj.name != BlenderUtils.DISPLAY_TABLE_NAME:
+            if (
+                obj.name not in no_save_objects
+                and obj.name != BlenderUtils.DISPLAY_TABLE_NAME
+            ):
                 obj_info = {
-                    "location": list(obj.location),
-                    "rotation_euler": list(obj.rotation_euler),
-                    "scale": list(obj.scale)
+                    "location": list(obj.matrix_world.translation),
+                    "rotation_euler": list(obj.matrix_world.to_euler()),
+                    "scale": list(obj.scale),
                 }
                 scene_info[obj.name] = obj_info
         scene_info[BlenderUtils.DISPLAY_TABLE_NAME] = display_table_config
@@ -365,5 +405,7 @@ class BlenderUtils:
         with open(scene_info_path, "w") as outfile:
             json.dump(scene_info, outfile)
             
-
-            
+    @staticmethod
+    def save_blend(scene_root_dir):
+        blend_path = os.path.join(scene_root_dir, "scene.blend")
+        bpy.ops.wm.save_as_mainfile(filepath=blend_path)

utils/cad_blender_util.py

@@ -0,0 +1,361 @@
+import os
+import json
+import bpy
+import time
+import gc
+import numpy as np
+import mathutils
+
+
+class CADBlenderUtils:
+
+    TABLE_NAME: str = "table"
+    CAMERA_NAME: str = "Camera"
+    CAMERA_RIGHT_NAME: str = "CameraRight"
+    CAMERA_OBJECT_NAME: str = "CameraObject"
+    DISPLAY_TABLE_NAME: str = "display_table"
+    MESH_FILE_NAME: str = "mesh.obj"
+
+    @staticmethod
+    def get_obj_path(obj_dir, name):
+        return os.path.join(obj_dir, name, CADBlenderUtils.MESH_FILE_NAME)
+
+    @staticmethod
+    def load_obj(name, mesh_path, scale=1):
+        print(mesh_path)
+        bpy.ops.wm.obj_import(filepath=mesh_path)
+        loaded_object = bpy.context.selected_objects[-1]
+        loaded_object.name = name
+        loaded_object.data.name = name
+        loaded_object.scale = (scale, scale, scale)
+        bpy.ops.rigidbody.object_add()
+        return loaded_object
+
+    @staticmethod
+    def get_obj(name):
+        return bpy.data.objects.get(name)
+
+    @staticmethod
+    def get_obj_pose(name):
+        obj = CADBlenderUtils.get_obj(name)
+        return np.asarray(obj.matrix_world)
+
+    @staticmethod
+    def add_plane(name, location, orientation, size=10):
+        bpy.ops.mesh.primitive_plane_add(size=size, location=location)
+        plane = bpy.context.selected_objects[-1]
+        plane.name = name
+        plane.rotation_euler = orientation
+        bpy.ops.rigidbody.object_add()
+        bpy.context.object.rigid_body.type = "PASSIVE"
+
+    @staticmethod
+    def setup_scene(init_light_and_camera_config, table_model_path, binocular_vision):
+        bpy.context.scene.render.engine = "BLENDER_EEVEE"
+        bpy.context.scene.display.shading.show_xray = False
+        bpy.context.scene.display.shading.use_dof = False
+        bpy.context.scene.display.render_aa = "OFF"
+        bpy.context.scene.view_settings.view_transform = "Standard"
+
+        bpy.context.scene.eevee.use_ssr = False  # 关闭屏幕空间反射
+        bpy.context.scene.eevee.use_bloom = False  # 关闭辉光
+        bpy.context.scene.eevee.use_gtao = False  # 关闭环境光遮蔽
+        bpy.context.scene.eevee.use_soft_shadows = False  # 关闭软阴影
+        bpy.context.scene.eevee.use_shadows = False  # 关闭所有阴影
+        bpy.context.scene.world.use_nodes = False  # 如果你不需要环境光，关闭环境节点
+
+        # bpy.context.scene.eevee.use_sss = False            # 关闭次表面散射
+
+        # 2. 设置最低的采样数
+        bpy.context.scene.eevee.taa_render_samples = 1
+        bpy.context.scene.eevee.taa_samples = 1
+        CADBlenderUtils.init_light_and_camera(
+            init_light_and_camera_config, binocular_vision
+        )
+
+    @staticmethod
+    def set_camera_params(camera, config, binocular_vision):
+
+        camera_object = bpy.data.objects.new(CADBlenderUtils.CAMERA_OBJECT_NAME, None)
+        bpy.context.collection.objects.link(camera_object)
+        cameras = [bpy.data.objects.get("Camera")]
+        camera.location = [0, 0, 0]
+        camera.rotation_euler = [0, 0, 0]
+        camera.parent = camera_object
+        if binocular_vision:
+            left_camera = cameras[0]
+            right_camera = left_camera.copy()
+            right_camera.name = CADBlenderUtils.CAMERA_RIGHT_NAME
+            right_camera.data = left_camera.data.copy()
+            right_camera.data.name = CADBlenderUtils.CAMERA_RIGHT_NAME
+            bpy.context.collection.objects.link(right_camera)
+            right_camera.parent = camera_object
+            right_camera.location = [config["eye_distance"] / 2, 0, 0]
+            left_camera.location = [-config["eye_distance"] / 2, 0, 0]
+            binocular_angle = config["eye_angle"]
+            half_angle = np.radians(binocular_angle / 2)
+
+            left_camera.rotation_euler[1] = -half_angle
+            right_camera.rotation_euler[1] = half_angle
+            cameras.append(right_camera)
+
+        for camera in cameras:
+            camera.data.clip_start = config["near_plane"]
+            camera.data.clip_end = config["far_plane"]
+
+            bpy.context.scene.render.resolution_x = config["resolution"][0]
+            bpy.context.scene.render.resolution_y = config["resolution"][1]
+            sensor_height = 24.0
+            focal_length = sensor_height / (
+                2 * np.tan(np.radians(config["fov_vertical"]) / 2)
+            )
+            camera.data.lens = focal_length
+            camera.data.sensor_width = (
+                sensor_height * config["resolution"][0] / config["resolution"][1]
+            )
+            camera.data.sensor_height = sensor_height
+
+    @staticmethod
+    def init_light_and_camera(init_light_and_camera_config, binocular_vision):
+
+        camera = CADBlenderUtils.get_obj(CADBlenderUtils.CAMERA_NAME)
+        CADBlenderUtils.set_camera_params(
+            camera,
+            init_light_and_camera_config[CADBlenderUtils.CAMERA_NAME],
+            binocular_vision,
+        )
+
+    @staticmethod
+    def get_obj_diag(name):
+        obj = CADBlenderUtils.get_obj(name)
+        return np.linalg.norm(obj.dimensions)
+
+    @staticmethod
+    def matrix_to_blender_pose(matrix):
+        location = matrix[:3, 3]
+        rotation_matrix = matrix[:3, :3]
+        rotation_matrix_blender = mathutils.Matrix(rotation_matrix.tolist())
+        rotation_euler = rotation_matrix_blender.to_euler()
+        return location, rotation_euler
+
+    @staticmethod
+    def set_camera_at(pose):
+        camera = CADBlenderUtils.get_obj(CADBlenderUtils.CAMERA_OBJECT_NAME)
+        location, rotation_euler = CADBlenderUtils.matrix_to_blender_pose(pose)
+
+        camera.location = location
+        camera.rotation_euler = rotation_euler
+
+    @staticmethod
+    def get_object_bottom_z(obj):
+        vertices = [v.co for v in obj.data.vertices]
+        vertices_world = [obj.matrix_world @ v for v in vertices]
+        min_z = min([v.z for v in vertices_world])
+        return min_z
+
+    @staticmethod
+    def render_normal_and_depth(
+        output_dir, file_name, binocular_vision=False, target_object=None
+    ):
+        # use pass z
+        bpy.context.scene.view_layers["ViewLayer"].use_pass_z = True
+        target_cameras = [CADBlenderUtils.CAMERA_NAME]
+        if binocular_vision:
+            target_cameras.append(CADBlenderUtils.CAMERA_RIGHT_NAME)
+            
+        for cam_name in target_cameras:
+            bpy.context.scene.camera = CADBlenderUtils.get_obj(cam_name)
+            cam_suffix = "L" if cam_name == CADBlenderUtils.CAMERA_NAME else "R"
+            scene = bpy.context.scene
+            scene.render.filepath = ""
+
+            mask_dir = os.path.join(output_dir, "normal")
+            if not os.path.exists(mask_dir):
+                os.makedirs(mask_dir)
+
+            scene.render.filepath = os.path.join(
+                output_dir, mask_dir, f"{file_name}_{cam_suffix}.exr"
+            ) 
+            
+            scene.render.image_settings.file_format = "OPEN_EXR"
+            scene.render.image_settings.color_mode = "RGB"
+            bpy.context.scene.view_settings.view_transform = "Raw"
+            scene.render.image_settings.color_depth = "16"
+            bpy.context.scene.render.filter_size = 1.5
+            scene.render.resolution_percentage = 100
+            scene.render.use_overwrite = False
+            scene.render.use_file_extension = False
+            scene.render.use_placeholder = False
+            scene.use_nodes = True
+            tree = scene.node_tree
+
+            for node in tree.nodes:
+                tree.nodes.remove(node)
+
+            rl = tree.nodes.new("CompositorNodeRLayers")
+
+            map_range = tree.nodes.new("CompositorNodeMapRange")
+            map_range.inputs["From Min"].default_value = 0.01
+            map_range.inputs["From Max"].default_value = 5
+            map_range.inputs["To Min"].default_value = 0
+            map_range.inputs["To Max"].default_value = 1
+            tree.links.new(rl.outputs["Depth"], map_range.inputs[0])
+
+            output_depth = tree.nodes.new("CompositorNodeOutputFile")
+
+            depth_dir = os.path.join(output_dir, "depth")
+            if not os.path.exists(depth_dir):
+                os.makedirs(depth_dir)
+            output_depth.base_path = depth_dir
+            output_depth.file_slots[0].path = f"{file_name}_{cam_suffix}.####"
+            output_depth.format.file_format = "PNG"
+            output_depth.format.color_mode = "BW"
+            output_depth.format.color_depth = "16"
+            tree.links.new(map_range.outputs[0], output_depth.inputs[0])
+            bpy.ops.render.render(write_still=True)
+
+        msg = "success"
+        return msg
+        
+    @staticmethod
+    def render_mask(
+        output_dir, file_name, binocular_vision=False, target_object=None
+    ):
+        target_cameras = [CADBlenderUtils.CAMERA_NAME]
+        if binocular_vision:
+            target_cameras.append(CADBlenderUtils.CAMERA_RIGHT_NAME)
+        
+        for cam_name in target_cameras:
+            bpy.context.scene.camera = CADBlenderUtils.get_obj(cam_name)
+            cam_suffix = "L" if cam_name == CADBlenderUtils.CAMERA_NAME else "R"
+            scene = bpy.context.scene
+            scene.render.filepath = ""
+
+            mask_dir = os.path.join(output_dir, "mask")
+            if not os.path.exists(mask_dir):
+                os.makedirs(mask_dir)
+
+            scene.render.filepath = os.path.join(
+                output_dir, mask_dir, f"{file_name}_{cam_suffix}.png"
+            )
+            scene.render.image_settings.color_depth = "8"
+            scene.render.resolution_percentage = 100
+            scene.render.use_overwrite = False
+            scene.render.use_file_extension = False
+            scene.render.use_placeholder = False
+
+            
+            bpy.ops.render.render(write_still=True)
+
+        msg = "success"
+        return msg
+
+    @staticmethod
+    def save_cam_params(scene_dir, idx, binocular_vision=False):
+        camera = CADBlenderUtils.get_obj(CADBlenderUtils.CAMERA_NAME)
+        extrinsic = np.array(camera.matrix_world)
+        cam_data = camera.data
+        focal_length = cam_data.lens
+        sensor_width = cam_data.sensor_width
+        sensor_height = cam_data.sensor_height
+        resolution_x = bpy.context.scene.render.resolution_x
+        resolution_y = bpy.context.scene.render.resolution_y
+        intrinsic = np.zeros((3, 3))
+        intrinsic[0, 0] = focal_length * resolution_x / sensor_width  # fx
+        intrinsic[1, 1] = focal_length * resolution_y / sensor_height  # fy
+        intrinsic[0, 2] = resolution_x / 2.0  # cx
+        intrinsic[1, 2] = resolution_y / 2.0  # cy
+        intrinsic[2, 2] = 1.0
+        cam_object = CADBlenderUtils.get_obj(CADBlenderUtils.CAMERA_OBJECT_NAME)
+        extrinsic_cam_object = np.array(cam_object.matrix_world)
+        data = {
+            "extrinsic": extrinsic.tolist(),
+            "extrinsic_cam_object": extrinsic_cam_object.tolist(),
+            "intrinsic": intrinsic.tolist(),
+            "far_plane": camera.data.clip_end,
+            "near_plane": camera.data.clip_start,
+        }
+        if binocular_vision:
+            right_camera = CADBlenderUtils.get_obj(CADBlenderUtils.CAMERA_RIGHT_NAME)
+            extrinsic_right = np.array(right_camera.matrix_world)
+            print("result:", extrinsic_right)
+
+            data["extrinsic_R"] = extrinsic_right.tolist()
+
+        cam_params_dir = os.path.join(scene_dir, "camera_params")
+        if not os.path.exists(cam_params_dir):
+            os.makedirs(cam_params_dir)
+        cam_params_path = os.path.join(cam_params_dir, f"{idx}.json")
+        with open(cam_params_path, "w") as f:
+            json.dump(data, f, indent=4)
+
+    @staticmethod
+    def reset_objects_and_platform():
+        all_objects = bpy.data.objects
+        keep_objects = {
+            "plane_floor",
+            "plane_ceil",
+            "plane_wall_1",
+            "plane_wall_2",
+            "plane_wall_3",
+            "plane_wall_4",
+        }
+        keep_objects.add(CADBlenderUtils.CAMERA_OBJECT_NAME)
+        keep_objects.add(CADBlenderUtils.CAMERA_NAME)
+        keep_objects.add(CADBlenderUtils.CAMERA_RIGHT_NAME)
+        keep_objects.add(CADBlenderUtils.TABLE_NAME)
+
+        for obj in all_objects:
+            if obj.name not in keep_objects:
+                bpy.data.objects.remove(obj, do_unlink=True)
+
+        for block in bpy.data.meshes:
+            if block.users == 0:
+                bpy.data.meshes.remove(block)
+        for block in bpy.data.materials:
+            if block.users == 0:
+                bpy.data.materials.remove(block)
+        for block in bpy.data.images:
+            if block.users == 0:
+                bpy.data.images.remove(block)
+
+        gc.collect()
+        bpy.context.scene.frame_set(0)
+
+    @staticmethod
+    def save_scene_info(scene_root_dir, display_table_config, target_name):
+        all_objects = bpy.data.objects
+        no_save_objects = {
+            "plane_floor",
+            "plane_ceil",
+            "plane_wall_1",
+            "plane_wall_2",
+            "plane_wall_3",
+            "plane_wall_4",
+        }
+        no_save_objects.add(CADBlenderUtils.CAMERA_OBJECT_NAME)
+        no_save_objects.add(CADBlenderUtils.CAMERA_NAME)
+        no_save_objects.add(CADBlenderUtils.CAMERA_RIGHT_NAME)
+        no_save_objects.add(CADBlenderUtils.TABLE_NAME)
+        scene_info = {}
+        for obj in all_objects:
+            if (
+                obj.name not in no_save_objects
+                and obj.name != CADBlenderUtils.DISPLAY_TABLE_NAME
+            ):
+                obj_info = {
+                    "location": list(obj.location),
+                    "rotation_euler": list(obj.rotation_euler),
+                    "scale": list(obj.scale),
+                }
+                scene_info[obj.name] = obj_info
+        scene_info[CADBlenderUtils.DISPLAY_TABLE_NAME] = display_table_config
+        scene_info["target_name"] = target_name
+        scene_info_path = os.path.join(scene_root_dir, "scene_info.json")
+        with open(scene_info_path, "w") as outfile:
+            json.dump(scene_info, outfile)
+            
+    @staticmethod
+    def save_blend(scene_root_dir):
+        blend_path = os.path.join(scene_root_dir, "scene.blend")
+        bpy.ops.wm.save_as_mainfile(filepath=blend_path)

utils/cad_view_sample_util.py

@@ -0,0 +1,146 @@
+
+import numpy as np
+import bmesh
+from collections import defaultdict
+from scipy.spatial.transform import Rotation as R
+from utils.pose import PoseUtil
+from utils.pts import PtsUtil
+import random
+
+class CADViewSampleUtil:
+    @staticmethod
+    def farthest_point_sampling(points, num_samples):
+        num_points = points.shape[0]
+        if num_samples >= num_points:
+            return points, np.arange(num_points)
+        sampled_indices = np.zeros(num_samples, dtype=int)
+        sampled_indices[0] = np.random.randint(num_points)
+        min_distances = np.full(num_points, np.inf)
+        for i in range(1, num_samples):
+            current_point = points[sampled_indices[i - 1]]
+            dist_to_current_point = np.linalg.norm(points - current_point, axis=1)
+            min_distances = np.minimum(min_distances, dist_to_current_point)
+            sampled_indices[i] = np.argmax(min_distances)
+        downsampled_points = points[sampled_indices]
+        return downsampled_points, sampled_indices
+    
+    @staticmethod
+    def voxel_downsample(points, voxel_size):
+        voxel_grid = defaultdict(list)
+        for i, point in enumerate(points):
+            voxel_index = tuple((point // voxel_size).astype(int))
+            voxel_grid[voxel_index].append(i)
+
+        downsampled_points = []
+        downsampled_indices = []
+        for indices in voxel_grid.values():
+            selected_index = indices[0]
+            downsampled_points.append(points[selected_index])
+            downsampled_indices.append(selected_index)
+
+        return np.array(downsampled_points), downsampled_indices
+
+    @staticmethod
+    def sample_view_data(obj, distance_range:tuple = (0.25,0.5), voxel_size:float = 0.005, max_views: int = 1, pertube_repeat:int = 1) -> dict:
+        view_data = {
+            "look_at_points": [],
+            "cam_positions": [],
+        }
+        mesh = obj.data
+        bm = bmesh.new()
+        bm.from_mesh(mesh)
+        bm.verts.ensure_lookup_table()
+        bm.faces.ensure_lookup_table()
+        bm.normal_update()
+        
+        look_at_points = []
+        cam_positions = []
+        normals = []
+        for v in bm.verts:
+            look_at_point = np.array(v.co)
+            
+            view_data["look_at_points"].append(look_at_point)
+            normal = np.zeros(3)
+            for loop in v.link_loops:
+                normal += np.array(loop.calc_normal())
+            normal /= len(v.link_loops)
+            normal = normal / np.linalg.norm(normal)
+            if np.isnan(normal).any():
+                continue
+            if np.dot(normal, look_at_point) < 0:
+                normal = -normal
+            normals.append(normal)
+            
+            for _ in range(pertube_repeat):
+                perturb_angle = np.radians(np.random.uniform(0, 10))
+                perturb_axis = np.random.normal(size=3)
+                perturb_axis /= np.linalg.norm(perturb_axis)
+                rotation_matrix = R.from_rotvec(perturb_angle * perturb_axis).as_matrix()
+                perturbed_normal = np.dot(rotation_matrix, normal)
+                middle_distance = (distance_range[0] + distance_range[1]) / 2
+                perturbed_distance = random.uniform(middle_distance-0.05, middle_distance+0.05)
+                cam_position = look_at_point + perturbed_distance * perturbed_normal
+                look_at_points.append(look_at_point)
+                cam_positions.append(cam_position)
+            
+            
+        bm.free() 
+        look_at_points = np.array(look_at_points)
+        cam_positions = np.array(cam_positions)
+        voxel_downsampled_look_at_points, selected_indices = CADViewSampleUtil.voxel_downsample(look_at_points, voxel_size)
+        voxel_downsampled_cam_positions = cam_positions[selected_indices]
+        voxel_downsampled_normals = np.array(normals)[selected_indices]
+
+        fps_downsampled_look_at_points, selected_indices = CADViewSampleUtil.farthest_point_sampling(voxel_downsampled_look_at_points, max_views*2)
+        fps_downsampled_cam_positions = voxel_downsampled_cam_positions[selected_indices]
+
+        view_data["look_at_points"] = fps_downsampled_look_at_points.tolist()
+        view_data["cam_positions"] = fps_downsampled_cam_positions.tolist()
+        view_data["normals"] = voxel_downsampled_normals
+        view_data["voxel_down_sampled_points"] = voxel_downsampled_look_at_points 
+        return view_data
+    
+    @staticmethod
+    def get_world_points_and_normals(view_data: dict, obj_world_pose: np.ndarray) -> tuple:
+        world_points = []
+        world_normals = []
+        for voxel_down_sampled_points, normal in zip(view_data["voxel_down_sampled_points"], view_data["normals"]):
+            voxel_down_sampled_points_world = obj_world_pose @ np.append(voxel_down_sampled_points, 1.0)
+            normal_world = obj_world_pose[:3, :3] @ normal
+            world_points.append(voxel_down_sampled_points_world[:3])
+            world_normals.append(normal_world)
+        return np.array(world_points), np.array(world_normals)
+    
+    @staticmethod
+    def get_cam_pose(view_data: dict, obj_world_pose: np.ndarray, max_views: int) -> np.ndarray:
+        cam_poses = []
+
+        for look_at_point, cam_position in zip(view_data["look_at_points"], view_data["cam_positions"]):
+            look_at_point_world = obj_world_pose @ np.append(look_at_point, 1.0)
+            cam_position_world = obj_world_pose @ np.append(cam_position, 1.0)
+            look_at_point_world = look_at_point_world[:3]  
+            cam_position_world = cam_position_world[:3]    
+            forward_vector = cam_position_world - look_at_point_world
+            forward_vector /= np.linalg.norm(forward_vector)
+            up_vector = np.array([0, 0, 1])
+            right_vector = np.cross(up_vector, forward_vector)
+            rotation_matrix = np.array([right_vector, up_vector, forward_vector]).T
+            cam_pose = np.eye(4)
+            cam_pose[:3, :3] = rotation_matrix
+            cam_pose[:3, 3] = cam_position_world
+            cam_poses.append(cam_pose)
+        if len(cam_poses) > max_views:
+            cam_points = np.array([cam_pose[:3, 3] for cam_pose in cam_poses])
+            _, indices = PtsUtil.fps_downsample_point_cloud(cam_points, max_views, require_idx=True)
+            cam_poses = [cam_poses[i] for i in indices]
+        
+        return np.array(cam_poses)
+        
+    @staticmethod
+    def sample_view_data_world_space(obj, distance_range:tuple = (0.3,0.5), voxel_size:float = 0.005, max_views: int=1, min_cam_table_included_degree:int=20, random_view_ratio:float = 0.2) -> dict:
+        obj_world_pose = np.asarray(obj.matrix_world)
+        view_data = CADViewSampleUtil.sample_view_data(obj, distance_range, voxel_size, max_views)
+        view_data["cam_poses"] = CADViewSampleUtil.get_cam_pose(view_data, obj_world_pose, max_views, min_cam_table_included_degree, random_view_ratio)
+        view_data["voxel_down_sampled_points"], view_data["normals"] =  CADViewSampleUtil.get_world_points_and_normals(view_data, obj_world_pose)
+        return view_data
+    

utils/material_util.py

@@ -0,0 +1,96 @@
+import bpy
+
+class MaterialUtil:
+
+    ''' --------- Basic --------- '''
+    @staticmethod
+    def change_object_material(obj, mat):
+        if obj.data.materials:
+            obj.data.materials[0] = mat
+        else:
+            obj.data.materials.append(mat)
+    
+    ''' ------- Materials ------- '''
+    @staticmethod
+    def create_normal_material():
+        normal_mat = bpy.data.materials.new(name="NormalMaterial")
+        normal_mat.use_nodes = True
+
+        nodes = normal_mat.node_tree.nodes
+        links = normal_mat.node_tree.links
+
+        nodes.clear()
+
+        geometry_node = nodes.new(type="ShaderNodeNewGeometry")
+        vector_transform_node = nodes.new(type="ShaderNodeVectorTransform")
+        separate_xyz_node = nodes.new(type="ShaderNodeSeparateXYZ")
+        multiply_node_x = nodes.new(type="ShaderNodeMath")
+        multiply_node_y = nodes.new(type="ShaderNodeMath")
+        multiply_node_z = nodes.new(type="ShaderNodeMath")
+        combine_xyz_node = nodes.new(type="ShaderNodeCombineXYZ")
+        light_path_node = nodes.new(type="ShaderNodeLightPath")
+        emission_node_1 = nodes.new(type="ShaderNodeEmission")
+        emission_node_2 = nodes.new(type="ShaderNodeEmission")
+        mix_shader_node_1 = nodes.new(type="ShaderNodeMixShader")
+        mix_shader_node_2 = nodes.new(type="ShaderNodeMixShader")
+        material_output_node = nodes.new(type="ShaderNodeOutputMaterial")
+
+        vector_transform_node.vector_type = 'VECTOR'
+        vector_transform_node.convert_from = 'WORLD'
+        vector_transform_node.convert_to = 'CAMERA'
+
+        multiply_node_x.operation = 'MULTIPLY'
+        multiply_node_x.inputs[1].default_value = 1.0
+
+        multiply_node_y.operation = 'MULTIPLY'
+        multiply_node_y.inputs[1].default_value = 1.0
+
+        multiply_node_z.operation = 'MULTIPLY'
+        multiply_node_z.inputs[1].default_value = -1.0
+
+        emission_node_1.inputs['Strength'].default_value = 1.0
+        emission_node_2.inputs['Strength'].default_value = 1.0
+
+        mix_shader_node_2.inputs['Fac'].default_value = 0.5
+
+        links.new(geometry_node.outputs['Normal'], vector_transform_node.inputs['Vector'])
+        links.new(vector_transform_node.outputs['Vector'], separate_xyz_node.inputs['Vector'])
+        links.new(separate_xyz_node.outputs['X'], multiply_node_x.inputs[0])
+        links.new(separate_xyz_node.outputs['Y'], multiply_node_y.inputs[0])
+        links.new(separate_xyz_node.outputs['Z'], multiply_node_z.inputs[0])
+        links.new(multiply_node_x.outputs['Value'], combine_xyz_node.inputs['X'])
+        links.new(multiply_node_y.outputs['Value'], combine_xyz_node.inputs['Y'])
+        links.new(multiply_node_z.outputs['Value'], combine_xyz_node.inputs['Z'])
+        links.new(combine_xyz_node.outputs['Vector'], emission_node_1.inputs['Color'])
+        links.new(light_path_node.outputs['Is Camera Ray'], mix_shader_node_1.inputs['Fac'])
+        links.new(emission_node_1.outputs['Emission'], mix_shader_node_1.inputs[2])
+        links.new(mix_shader_node_1.outputs['Shader'], mix_shader_node_2.inputs[1])
+        links.new(emission_node_2.outputs['Emission'], mix_shader_node_2.inputs[2])
+        links.new(mix_shader_node_2.outputs['Shader'], material_output_node.inputs['Surface'])
+        return normal_mat
+    
+    @staticmethod
+    def create_mask_material(color=(1.0, 1.0, 1.0)): 
+        mask_mat = bpy.data.materials.new(name="MaskMaterial")
+        mask_mat.use_nodes = True
+
+        nodes = mask_mat.node_tree.nodes
+        links = mask_mat.node_tree.links
+
+        nodes.clear()
+        emission_node = nodes.new(type="ShaderNodeEmission")
+        emission_node.inputs['Color'].default_value = (*color, 1.0) 
+        emission_node.inputs['Strength'].default_value = 1.0
+        material_output_node = nodes.new(type="ShaderNodeOutputMaterial")
+        links.new(emission_node.outputs['Emission'], material_output_node.inputs['Surface'])
+
+        return mask_mat
+
+    
+
+# -------- debug --------
+if __name__ == "__main__":
+    cube = bpy.data.objects.get("Cube")
+    normal_mat = MaterialUtil.create_normal_material()
+    MaterialUtil.change_object_material(cube, normal_mat)
+    

utils/pose.py

@@ -149,3 +149,18 @@ class PoseUtil:
         if debug:
             print("uniform scale:", scale)
         return scale
+
+    @staticmethod
+    def rotation_matrix_from_axis_angle(axis, angle):
+        cos_angle = np.cos(angle)
+        sin_angle = np.sin(angle)
+        one_minus_cos = 1 - cos_angle
+
+        x, y, z = axis
+        rotation_matrix = np.array([
+            [cos_angle + x*x*one_minus_cos, x*y*one_minus_cos - z*sin_angle, x*z*one_minus_cos + y*sin_angle],
+            [y*x*one_minus_cos + z*sin_angle, cos_angle + y*y*one_minus_cos, y*z*one_minus_cos - x*sin_angle],
+            [z*x*one_minus_cos - y*sin_angle, z*y*one_minus_cos + x*sin_angle, cos_angle + z*z*one_minus_cos]
+        ])
+
+        return rotation_matrix

utils/pts.py

@@ -0,0 +1,83 @@
+import numpy as np
+
+
+class PtsUtil:
+    
+    @staticmethod
+    def random_downsample_point_cloud(point_cloud, num_points, require_idx=False):
+        if point_cloud.shape[0] == 0:
+            if require_idx:
+                return point_cloud, np.array([])
+            return point_cloud
+        idx = np.random.choice(len(point_cloud), num_points, replace=True)
+        if require_idx:
+            return point_cloud[idx], idx
+        return point_cloud[idx]
+    
+    @staticmethod
+    def fps_downsample_point_cloud(point_cloud, num_points, require_idx=False):
+        N = point_cloud.shape[0]
+        mask = np.zeros(N, dtype=bool)
+
+        sampled_indices = np.zeros(num_points, dtype=int)
+        sampled_indices[0] = np.random.randint(0, N)
+        distances = np.linalg.norm(point_cloud - point_cloud[sampled_indices[0]], axis=1)
+        for i in range(1, num_points):
+            farthest_index = np.argmax(distances)
+            sampled_indices[i] = farthest_index
+            mask[farthest_index] = True
+
+            new_distances = np.linalg.norm(point_cloud - point_cloud[farthest_index], axis=1)
+            distances = np.minimum(distances, new_distances)
+        
+        sampled_points = point_cloud[sampled_indices]
+        if require_idx:
+            return sampled_points, sampled_indices
+        return sampled_points
+    
+    @staticmethod
+    def voxelize_points(points, voxel_size):
+        voxel_indices = np.floor(points / voxel_size).astype(np.int32)
+        unique_voxels = np.unique(voxel_indices, axis=0, return_inverse=True)
+        return unique_voxels
+
+    @staticmethod
+    def transform_point_cloud(points, pose_mat):
+        points_h = np.concatenate([points, np.ones((points.shape[0], 1))], axis=1)
+        points_h = np.dot(pose_mat, points_h.T).T
+        return points_h[:, :3]
+    
+    @staticmethod
+    def get_overlapping_points(point_cloud_L, point_cloud_R, voxel_size=0.005, require_idx=False):
+        voxels_L, indices_L = PtsUtil.voxelize_points(point_cloud_L, voxel_size)
+        voxels_R, _ = PtsUtil.voxelize_points(point_cloud_R, voxel_size)
+
+        voxel_indices_L = voxels_L.view([("", voxels_L.dtype)] * 3)
+        voxel_indices_R = voxels_R.view([("", voxels_R.dtype)] * 3)
+        overlapping_voxels = np.intersect1d(voxel_indices_L, voxel_indices_R)
+        mask_L = np.isin(
+            indices_L, np.where(np.isin(voxel_indices_L, overlapping_voxels))[0]
+        )
+        overlapping_points = point_cloud_L[mask_L]
+        if require_idx:
+            return overlapping_points, mask_L
+        return overlapping_points
+    
+    @staticmethod
+    def filter_points(points, normals, cam_pose, theta=45, z_range=(0.2, 0.45)):
+        
+        """ filter with normal """ 
+        normals_normalized = normals / np.linalg.norm(normals, axis=1, keepdims=True)
+        cos_theta = np.dot(normals_normalized, np.array([0, 0, 1]))
+        theta_rad = np.deg2rad(theta)
+        idx = cos_theta > np.cos(theta_rad)
+        filtered_sampled_points = points[idx] 
+        
+        
+        """ filter with z range """
+        points_cam = PtsUtil.transform_point_cloud(filtered_sampled_points, np.linalg.inv(cam_pose))
+        idx = (points_cam[:, 2] > z_range[0]) & (points_cam[:, 2] < z_range[1])
+        z_filtered_points = filtered_sampled_points[idx]
+        
+        return z_filtered_points[:, :3]
+        

utils/view_sample_util.py

@@ -3,7 +3,8 @@ import numpy as np
 import bmesh
 from collections import defaultdict
 from scipy.spatial.transform import Rotation as R
-from blender.pose import PoseUtil
+from utils.pose import PoseUtil
+from utils.pts import PtsUtil
 import random
 
 class ViewSampleUtil:
@@ -71,7 +72,7 @@ class ViewSampleUtil:
             normals.append(normal)
             
             for _ in range(pertube_repeat):
-                perturb_angle = np.radians(np.random.uniform(0, 30))
+                perturb_angle = np.radians(np.random.uniform(0, 10))
                 perturb_axis = np.random.normal(size=3)
                 perturb_axis /= np.linalg.norm(perturb_axis)
                 rotation_matrix = R.from_rotvec(perturb_angle * perturb_axis).as_matrix()
@@ -127,12 +128,33 @@ class ViewSampleUtil:
             
             up_vector = np.array([0, 0, 1])
 
+            dot_product = np.dot(forward_vector, up_vector)
+            angle = np.degrees(np.arccos(dot_product))
             right_vector = np.cross(up_vector, forward_vector)
-            right_vector /= np.linalg.norm(right_vector)
            
-            corrected_up_vector = np.cross(forward_vector, right_vector)
-            rotation_matrix = np.array([right_vector, corrected_up_vector, forward_vector]).T
+            if angle > 90 - min_cam_table_included_degree:
+                max_angle = 90 - min_cam_table_included_degree
+                min_angle = max(90 - min_cam_table_included_degree*2, 30)
+                target_angle = np.random.uniform(min_angle, max_angle)
+                angle_difference = np.radians(target_angle - angle)
 
+                rotation_axis = np.cross(forward_vector, up_vector)
+                rotation_axis /= np.linalg.norm(rotation_axis)
+                rotation_matrix = PoseUtil.rotation_matrix_from_axis_angle(rotation_axis, -angle_difference)
+                new_cam_position_world = np.dot(rotation_matrix, cam_position_world - look_at_point_world) + look_at_point_world
+                cam_position_world = new_cam_position_world
+                forward_vector = cam_position_world - look_at_point_world
+                forward_vector /= np.linalg.norm(forward_vector)
+                right_vector = np.cross(up_vector, forward_vector)
+                right_vector /= np.linalg.norm(right_vector)
+
+                corrected_up_vector = np.cross(forward_vector, right_vector)
+                rotation_matrix = np.array([right_vector, corrected_up_vector, forward_vector]).T
+            else:
+                right_vector = np.cross(up_vector, forward_vector)
+                right_vector /= np.linalg.norm(right_vector)
+                corrected_up_vector = np.cross(forward_vector, right_vector)
+                rotation_matrix = np.array([right_vector, corrected_up_vector, forward_vector]).T
             cam_pose = np.eye(4)
             cam_pose[:3, :3] = rotation_matrix
             cam_pose[:3, 3] = cam_position_world
@@ -153,7 +175,8 @@ class ViewSampleUtil:
                     filtered_cam_poses.append(pertube_pose @ cam_pose)
                     
         if len(filtered_cam_poses) > max_views:
-            indices = np.random.choice(len(filtered_cam_poses), max_views, replace=False)
+            cam_points = np.array([cam_pose[:3, 3] for cam_pose in filtered_cam_poses])
+            _, indices = PtsUtil.fps_downsample_point_cloud(cam_points, max_views, require_idx=True)
             filtered_cam_poses = [filtered_cam_poses[i] for i in indices]
             
         return np.array(filtered_cam_poses)