3d gaussian splatting核心代码注释（python部分）

train.py：

def training(dataset, opt, pipe, testing_iterations, saving_iterations, checkpoint_iterations, checkpoint, debug_from):
    first_iter = 0
    tb_writer = prepare_output_and_logger(dataset)
    gaussians = GaussianModel(dataset.sh_degree)         # 创建GaussianModel模型，给点云中的每个点去创建一个3D gaussian
    scene = Scene(dataset, gaussians)                    # 加载数据集和每张图片对应的camera的参数
    gaussians.training_setup(opt)
    if checkpoint:                  #如果checkpoint非空，则加载checkpoint，恢复模型参数和优化参数
        (model_params, first_iter) = torch.load(checkpoint)
        gaussians.restore(model_params, opt)
    #设置背景值，白色[1, 1, 1]，否则[0, 0, 0]
    bg_color = [1, 1, 1] if dataset.white_background else [0, 0, 0]
    background = torch.tensor(bg_color, dtype=torch.float32, device="cuda")
    #记录gpu运行时间
    iter_start = torch.cuda.Event(enable_timing = True)
    iter_end = torch.cuda.Event(enable_timing = True)

    ####################迭代训练开始###########################
    viewpoint_stack = None
    ema_loss_for_log = 0.0
    progress_bar = tqdm(range(first_iter, opt.iterations), desc="Training progress")
    first_iter += 1
    for iteration in range(first_iter, opt.iterations + 1):        
        if network_gui.conn == None:
            network_gui.try_connect()
        while network_gui.conn != None:
            try:
                net_image_bytes = None
                custom_cam, do_training, pipe.convert_SHs_python, pipe.compute_cov3D_python, keep_alive, scaling_modifer = network_gui.receive()
                if custom_cam != None:
                    net_image = render(custom_cam, gaussians, pipe, background, scaling_modifer)["render"]
                    net_image_bytes = memoryview((torch.clamp(net_image, min=0, max=1.0) * 255).byte().permute(1, 2, 0).contiguous().cpu().numpy())
                network_gui.send(net_image_bytes, dataset.source_path)
                if do_training and ((iteration < int(opt.iterations)) or not keep_alive):
                    break
            except Exception as e:
                network_gui.conn = None
        #记录开始时间
        iter_start.record()
        # 对xyz的学习率进行调整
        gaussians.update_learning_rate(iteration)

        # Every 1000 its we increase the levels of SH up to a maximum degree
        if iteration % 1000 == 0:
            gaussians.oneupSHdegree()

        # Pick a random Camera  #随机选择一个图片及其相应的相机视角(内外参)
        if not viewpoint_stack:
            viewpoint_stack = scene.getTrainCameras().copy()
        viewpoint_cam = viewpoint_stack.pop(randint(0, len(viewpoint_stack)-1))

        # Render，在debug_from步时，启动调试
        if (iteration - 1) == debug_from:
            pipe.debug = True
        #随机背景与否
        bg = torch.rand((3), device="cuda") if opt.random_background else background
        #render 函数参数一：上步随机选择的相机，参数二：高斯模型，参数三：管道参数，参数四：背景
        render_pkg = render(viewpoint_cam, gaussians, pipe, bg)
        image, viewspace_point_tensor, visibility_filter, radii = render_pkg["render"], render_pkg["viewspace_points"], render_pkg["visibility_filter"], render_pkg["radii"]

        # Loss
        gt_image = viewpoint_cam.original_image.cuda()
        Ll1 = l1_loss(image, gt_image)
        loss = (1.0 - opt.lambda_dssim) * Ll1 + opt.lambda_dssim * (1.0 - ssim(image, gt_image))


        loss.backward()
        #记录结束时间
        iter_end.record()

        with torch.no_grad():
            # Progress bar更新进度条，每10步更新一下后缀内容及进度
            ema_loss_for_log = 0.4 * loss.item() + 0.6 * ema_loss_for_log
            if iteration % 10 == 0:
                progress_bar.set_postfix({"Loss": f"{ema_loss_for_log:.{7}f}"})
                progress_bar.update(10)
            if iteration == opt.iterations:
                progress_bar.close()

            # Log and save
            training_report(tb_writer, iteration, Ll1, loss, l1_loss, iter_start.elapsed_time(iter_end), testing_iterations, scene, render, (pipe, background))
            if (iteration in saving_iterations):
                print("\n[ITER {}] Saving Gaussians".format(iteration))
                scene.save(iteration)

            # Densification
            if iteration < opt.densify_until_iter:#15000步以前
                # Keep track of max radii in image-space for pruning
                gaussians.max_radii2D[visibility_filter] = torch.max(gaussians.max_radii2D[visibility_filter], radii[visibility_filter])
                # 统计3D gaussian均值(xyz)的梯度, 用于对3D gaussians的克隆或者切分
                gaussians.add_densification_stats(viewspace_point_tensor, visibility_filter)
                #每100步进行下面的操作，对3D gaussians进行克隆或者切分, 并将opacity小于一定阈值的3D gaussians进行删除
                if iteration > opt.densify_from_iter and iteration % opt.densification_interval == 0:
                    size_threshold = 20 if iteration > opt.opacity_reset_interval else None
                    gaussians.densify_and_prune(opt.densify_grad_threshold, 0.005, scene.cameras_extent, size_threshold)
                #每3000步重设不透明度
                if iteration % opt.opacity_reset_interval == 0 or (dataset.white_background and iteration == opt.densify_from_iter):
                    gaussians.reset_opacity()

            # Optimizer step
            if iteration < opt.iterations:#30000步以前
                gaussians.optimizer.step()
                gaussians.optimizer.zero_grad(set_to_none = True)

            if (iteration in checkpoint_iterations):
                print("\n[ITER {}] Saving Checkpoint".format(iteration))
                torch.save((gaussians.capture(), iteration), scene.model_path + "/chkpnt" + str(iteration) + ".pth")
        ####################迭代训练结束###########################
def prepare_output_and_logger(args):    
    if not args.model_path:
        if os.getenv('OAR_JOB_ID'):
            unique_str=os.getenv('OAR_JOB_ID')
        else:
            unique_str = str(uuid.uuid4())
        #生成一个随机的UUID，并取前十位作为输出目录名称
        args.model_path = os.path.join("./output/", unique_str[0:10])
        
    # Set up output folder
    print("Output folder: {}".format(args.model_path))
    os.makedirs(args.model_path, exist_ok = True)
    #配置参数写入到cfg_args文件中
    with open(os.path.join(args.model_path, "cfg_args"), 'w') as cfg_log_f:
        cfg_log_f.write(str(Namespace(**vars(args))))

    # Create Tensorboard writer
    tb_writer = None
    if TENSORBOARD_FOUND:
        tb_writer = SummaryWriter(args.model_path)
    else:
        print("Tensorboard not available: not logging progress")
    return tb_writer
#写入日志
def training_report(tb_writer, iteration, Ll1, loss, l1_loss, elapsed, testing_iterations, scene : Scene, renderFunc, renderArgs):
    if tb_writer:
        tb_writer.add_scalar('train_loss_patches/l1_loss', Ll1.item(), iteration)
        tb_writer.add_scalar('train_loss_patches/total_loss', loss.item(), iteration)
        tb_writer.add_scalar('iter_time', elapsed, iteration)

    # Report test and samples of training set
    if iteration in testing_iterations:
        torch.cuda.empty_cache()
        validation_configs = ({'name': 'test', 'cameras' : scene.getTestCameras()}, 
                              {'name': 'train', 'cameras' : [scene.getTrainCameras()[idx % len(scene.getTrainCameras())] for idx in range(5, 30, 5)]})

        for config in validation_configs:
            if config['cameras'] and len(config['cameras']) > 0:
                l1_test = 0.0
                psnr_test = 0.0
                for idx, viewpoint in enumerate(config['cameras']):
                    image = torch.clamp(renderFunc(viewpoint, scene.gaussians, *renderArgs)["render"], 0.0, 1.0)
                    gt_image = torch.clamp(viewpoint.original_image.to("cuda"), 0.0, 1.0)
                    if tb_writer and (idx < 5):
                        tb_writer.add_images(config['name'] + "_view_{}/render".format(viewpoint.image_name), image[None], global_step=iteration)
                        if iteration == testing_iterations[0]:
                            tb_writer.add_images(config['name'] + "_view_{}/ground_truth".format(viewpoint.image_name), gt_image[None], global_step=iteration)
                    l1_test += l1_loss(image, gt_image).mean().double()
                    psnr_test += psnr(image, gt_image).mean().double()
                psnr_test /= len(config['cameras'])
                l1_test /= len(config['cameras'])          
                print("\n[ITER {}] Evaluating {}: L1 {} PSNR {}".format(iteration, config['name'], l1_test, psnr_test))
                if tb_writer:
                    tb_writer.add_scalar(config['name'] + '/loss_viewpoint - l1_loss', l1_test, iteration)
                    tb_writer.add_scalar(config['name'] + '/loss_viewpoint - psnr', psnr_test, iteration)

        if tb_writer:
            tb_writer.add_histogram("scene/opacity_histogram", scene.gaussians.get_opacity, iteration)
            tb_writer.add_scalar('total_points', scene.gaussians.get_xyz.shape[0], iteration)
        torch.cuda.empty_cache()

if __name__ == "__main__":
    # Set up command line argument parser
    parser = ArgumentParser(description="Training script parameters")
    lp = ModelParams(parser)
    op = OptimizationParams(parser)
    pp = PipelineParams(parser)
    parser.add_argument('--ip', type=str, default="127.0.0.1")
    parser.add_argument('--port', type=int, default=6009)
    parser.add_argument('--debug_from', type=int, default=-1)
    parser.add_argument('--detect_anomaly', action='store_true', default=False)
    parser.add_argument("--test_iterations", nargs="+", type=int, default=[7_000, 30_000])
    parser.add_argument("--save_iterations", nargs="+", type=int, default=[7_000, 30_000])
    parser.add_argument("--quiet", action="store_true")
    parser.add_argument("--checkpoint_iterations", nargs="+", type=int, default=[])
    parser.add_argument("--start_checkpoint", type=str, default = None)
    args = parser.parse_args(sys.argv[1:])
    args.save_iterations.append(args.iterations)
    
    print("Optimizing " + args.model_path)

    # Initialize system state (RNG 随机数生成器)
    safe_state(args.quiet)

    # Start GUI server, configure and run training
    # 在ip的port端口监听
    network_gui.init(args.ip, args.port)
    # 正向传播时：关闭自动求导的异常侦测
    torch.autograd.set_detect_anomaly(args.detect_anomaly)
    #开始训练，函数参数：模型参数lp，优化参数，管道参数，测试迭代值，保存迭代值，检查点迭代值，开始检查点，调试来源。
    training(lp.extract(args), op.extract(args), pp.extract(args), args.test_iterations, args.save_iterations, args.checkpoint_iterations, args.start_checkpoint, args.debug_from)
    #结束
    # All done
    print("\nTraining complete.")

scene/__init__.py

class Scene:

    gaussians : GaussianModel
    #场景初始化，参数：模型参数和高斯模型
    def __init__(self, args : ModelParams, gaussians : GaussianModel, load_iteration=None, shuffle=True, resolution_scales=[1.0]):
        """b
        :param path: Path to colmap scene main folder.
        """
        self.model_path = args.model_path
        self.loaded_iter = None
        self.gaussians = gaussians
        # 是否加载迭代,load_iteration是迭代值
        # 寻找是否有训练过的记录, 如果没有则为初次训练, 需要从COLMAP创建的点云中初始化每个点对应的3D gaussian
        # 以及将每张图片对应的相机参数dump到`cameras.json`文件中
        if load_iteration:
            if load_iteration == -1:
                self.loaded_iter = searchForMaxIteration(os.path.join(self.model_path, "point_cloud"))
            else:
                self.loaded_iter = load_iteration
            print("Loading trained model at iteration {}".format(self.loaded_iter))

        self.train_cameras = {}
        self.test_cameras = {}
        # 从COLMAP或Blender中读取每张图片, 以及每张图片对应的相机内外参
        # 如果源目录中有sparse目录，则读取数据得到场景信息scene_info
        if os.path.exists(os.path.join(args.source_path, "sparse")):
            scene_info = sceneLoadTypeCallbacks["Colmap"](args.source_path, args.images, args.eval)
        #如果有transforms_train.json文件，则调用readNerfSyntheticInfo
        elif os.path.exists(os.path.join(args.source_path, "transforms_train.json")):
            print("Found transforms_train.json file, assuming Blender data set!")
            scene_info = sceneLoadTypeCallbacks["Blender"](args.source_path, args.white_background, args.eval)
        else:
            assert False, "Could not recognize scene type!"
        # 如果loaded_iter为空（没有设置从已经训练过的位置开始训练）
        # 将每张图片对应的相机参数dump到`cameras.json`文件中
        if not self.loaded_iter:
            #拷贝ply文件到输出目录，命名为input.ply
            with open(scene_info.ply_path, 'rb') as src_file, open(os.path.join(self.model_path, "input.ply") , 'wb') as dest_file:
                dest_file.write(src_file.read())
            json_cams = []
            #camlist追加scene_info的测试集和训练集
            camlist = []
            if scene_info.test_cameras:
                camlist.extend(scene_info.test_cameras)
            if scene_info.train_cameras:
                camlist.extend(scene_info.train_cameras)
            #把camlist的数据转成json存到cameras.json文件中
            for id, cam in enumerate(camlist):
                json_cams.append(camera_to_JSON(id, cam))
            with open(os.path.join(self.model_path, "cameras.json"), 'w') as file:
                json.dump(json_cams, file)
        # 随机打乱所有图片和对应相机的顺序，打乱train_cameras
        if shuffle:
            random.shuffle(scene_info.train_cameras)  # Multi-res consistent random shuffling
            random.shuffle(scene_info.test_cameras)  # Multi-res consistent random shuffling
        # 把getNerfppNorm返回的结果的半径赋给cameras_extent，所有相机的中心点位置到最远camera的距离
        self.cameras_extent = scene_info.nerf_normalization["radius"]
        #对每一个分辨率缩放，计算cameraList
        for resolution_scale in resolution_scales:
            print("Loading Training Cameras")
            self.train_cameras[resolution_scale] = cameraList_from_camInfos(scene_info.train_cameras, resolution_scale, args)
            print("Loading Test Cameras")
            self.test_cameras[resolution_scale] = cameraList_from_camInfos(scene_info.test_cameras, resolution_scale, args)
        # 如果是初次训练, 则从COLMAP创建的点云中初始化每个点对应的3D gaussian, 否则直接从之前保存的模型文件中读取3D gaussian
        if self.loaded_iter:
            self.gaussians.load_ply(os.path.join(self.model_path,
                                                           "point_cloud",
                                                           "iteration_" + str(self.loaded_iter),
                                                           "point_cloud.ply"))
        else:
            self.gaussians.create_from_pcd(scene_info.point_cloud, self.cameras_extent)

scene/gaussian_model.py

class GaussianModel:
    def setup_functions(self):
        # 从尺度和旋转参数中去构建3Dgaussian的协方差矩阵，L=RS，L@L的逆等于RSStRt，再去除对称性得到6个值表示对称的协方差矩阵
        def build_covariance_from_scaling_rotation(scaling, scaling_modifier, rotation):
            L = build_scaling_rotation(scaling_modifier * scaling, rotation)
            actual_covariance = L @ L.transpose(1, 2)
            symm = strip_symmetric(actual_covariance)
            return symm
        
        self.scaling_activation = torch.exp                     # 将尺度限制为非负数
        self.scaling_inverse_activation = torch.log

        self.covariance_activation = build_covariance_from_scaling_rotation

        self.opacity_activation = torch.sigmoid                 # 将不透明度限制在0-1的范围内
        self.inverse_opacity_activation = inverse_sigmoid

        self.rotation_activation = torch.nn.functional.normalize


    def __init__(self, sh_degree : int):
        self.active_sh_degree = 0
        self.max_sh_degree = sh_degree
        self._xyz = torch.empty(0)                  # 中心点位置, 也即3Dgaussian的均值
        self._features_dc = torch.empty(0)          # 第一个球谐系数, 球谐系数用来表示RGB颜色
        self._features_rest = torch.empty(0)        # 其余球谐系数
        self._scaling = torch.empty(0)              # 尺度
        self._rotation = torch.empty(0)             # 旋转参数, 四元组
        self._opacity = torch.empty(0)              # 不透明度
        self.max_radii2D = torch.empty(0)           # 投影到2D时, 每个2D gaussian最大的半径
        self.xyz_gradient_accum = torch.empty(0)    # 3Dgaussian的均值的累积梯度
        self.denom = torch.empty(0)
        self.optimizer = None
        self.percent_dense = 0
        self.spatial_lr_scale = 0
        self.setup_functions()

    def capture(self):
        return (
            self.active_sh_degree,
            self._xyz,
            self._features_dc,
            self._features_rest,
            self._scaling,
            self._rotation,
            self._opacity,
            self.max_radii2D,
            self.xyz_gradient_accum,
            self.denom,
            self.optimizer.state_dict(),
            self.spatial_lr_scale,
        )
    
    def restore(self, model_args, training_args):
        (self.active_sh_degree, 
        self._xyz, 
        self._features_dc, 
        self._features_rest,
        self._scaling, 
        self._rotation, 
        self._opacity,
        self.max_radii2D, 
        xyz_gradient_accum, 
        denom,
        opt_dict, 
        self.spatial_lr_scale) = model_args
        self.training_setup(training_args)
        self.xyz_gradient_accum = xyz_gradient_accum
        self.denom = denom
        self.optimizer.load_state_dict(opt_dict)

    @property
    def get_scaling(self):
        return self.scaling_activation(self._scaling)
    
    @property
    def get_rotation(self):
        return self.rotation_activation(self._rotation)
    
    @property
    def get_xyz(self):
        return self._xyz
    
    @property
    def get_features(self):
        features_dc = self._features_dc
        features_rest = self._features_rest
        return torch.cat((features_dc, features_rest), dim=1)
    
    @property
    def get_opacity(self):
        return self.opacity_activation(self._opacity)
    
    def get_covariance(self, scaling_modifier = 1):
        return self.covariance_activation(self.get_scaling, scaling_modifier, self._rotation)

    def oneupSHdegree(self):
        if self.active_sh_degree < self.max_sh_degree:
            self.active_sh_degree += 1
    #参数一：稀疏点云，参数二：球半径
    def create_from_pcd(self, pcd : BasicPointCloud, spatial_lr_scale : float):
        self.spatial_lr_scale = spatial_lr_scale
        #点云转tensor类型，送到gpu
        fused_point_cloud = torch.tensor(np.asarray(pcd.points)).float().cuda()
        #把点云的颜色转成sh系数，送到gpu
        fused_color = RGB2SH(torch.tensor(np.asarray(pcd.colors)).float().cuda())
        #特征长度为16，第一个维度为sh系数，其他赋0
        features = torch.zeros((fused_color.shape[0], 3, (self.max_sh_degree + 1) ** 2)).float().cuda()
        features[:, :3, 0 ] = fused_color
        features[:, 3:, 1:] = 0.0

        print("Number of points at initialisation : ", fused_point_cloud.shape[0])
        #调用simple_knn的distCUDA2函数，计算点云中的每个点到与其最近的K个点的平均距离的平方
        dist2 = torch.clamp_min(distCUDA2(torch.from_numpy(np.asarray(pcd.points)).float().cuda()), 0.0000001)
        #距离值取对数，得到缩放值，repeat变成3个
        scales = torch.log(torch.sqrt(dist2))[...,None].repeat(1, 3)
        rots = torch.zeros((fused_point_cloud.shape[0], 4), device="cuda")
        rots[:, 0] = 1
        #0.1的逆sigmoid值-2.1976，也即是-2.1976的sigmoid值0.1，初始化设置为不透明度
        opacities = inverse_sigmoid(0.1 * torch.ones((fused_point_cloud.shape[0], 1), dtype=torch.float, device="cuda"))
        #初始化位置，sh系数（直接+剩余），缩放（3个轴），旋转（四元数），不透明度（逆sigmoid的值），
        self._xyz = nn.Parameter(fused_point_cloud.requires_grad_(True))
        self._features_dc = nn.Parameter(features[:,:,0:1].transpose(1, 2).contiguous().requires_grad_(True))
        self._features_rest = nn.Parameter(features[:,:,1:].transpose(1, 2).contiguous().requires_grad_(True))
        self._scaling = nn.Parameter(scales.requires_grad_(True))
        self._rotation = nn.Parameter(rots.requires_grad_(True))
        self._opacity = nn.Parameter(opacities.requires_grad_(True))
        self.max_radii2D = torch.zeros((self.get_xyz.shape[0]), device="cuda")

    #
    def training_setup(self, training_args):
        self.percent_dense = training_args.percent_dense
        self.xyz_gradient_accum = torch.zeros((self.get_xyz.shape[0], 1), device="cuda")
        self.denom = torch.zeros((self.get_xyz.shape[0], 1), device="cuda")
        #参数列表，每一项是一个字典，包含位置、f_dc、f_rest、不透明度、缩放、旋转
        l = [
            {'params': [self._xyz], 'lr': training_args.position_lr_init * self.spatial_lr_scale, "name": "xyz"},
            {'params': [self._features_dc], 'lr': training_args.feature_lr, "name": "f_dc"},
            {'params': [self._features_rest], 'lr': training_args.feature_lr / 20.0, "name": "f_rest"},
            {'params': [self._opacity], 'lr': training_args.opacity_lr, "name": "opacity"},
            {'params': [self._scaling], 'lr': training_args.scaling_lr, "name": "scaling"},
            {'params': [self._rotation], 'lr': training_args.rotation_lr, "name": "rotation"}
        ]
        # 创建optimizer
        self.optimizer = torch.optim.Adam(l, lr=0.0, eps=1e-15)
        # 创建对xyz参数进行学习率调整的scheduler
        self.xyz_scheduler_args = get_expon_lr_func(lr_init=training_args.position_lr_init*self.spatial_lr_scale,
                                                    lr_final=training_args.position_lr_final*self.spatial_lr_scale,
                                                    lr_delay_mult=training_args.position_lr_delay_mult,
                                                    max_steps=training_args.position_lr_max_steps)

scene/dataset_readers.py

class SceneInfo(NamedTuple):
    point_cloud: BasicPointCloud
    train_cameras: list
    test_cameras: list
    nerf_normalization: dict
    ply_path: str
#计算这些相机的平均中心的负值和所在球半径大小
def getNerfppNorm(cam_info):
    #计算这些相机的平均中心，以及到平均中心的最远距离，返回平均中心和最远距离
    def get_center_and_diag(cam_centers):
        cam_centers = np.hstack(cam_centers)
        avg_cam_center = np.mean(cam_centers, axis=1, keepdims=True)
        center = avg_cam_center
        dist = np.linalg.norm(cam_centers - center, axis=0, keepdims=True)
        diagonal = np.max(dist)
        return center.flatten(), diagonal

    cam_centers = []
    #计算相机在世界坐标系的位置，并放入cam_centers列表中
    for cam in cam_info:
        W2C = getWorld2View2(cam.R, cam.T)
        C2W = np.linalg.inv(W2C)
        cam_centers.append(C2W[:3, 3:4])

    center, diagonal = get_center_and_diag(cam_centers)
    radius = diagonal * 1.1

    translate = -center

    return {"translate": translate, "radius": radius}
#读取相机信息，参数：相机外参，内参，图片目录，输出到列表中，列表每一项是CameraInfo类
def readColmapCameras(cam_extrinsics, cam_intrinsics, images_folder):
    cam_infos = []
    for idx, key in enumerate(cam_extrinsics):
        sys.stdout.write('\r')
        # the exact output you're looking for:
        sys.stdout.write("Reading camera {}/{}".format(idx+1, len(cam_extrinsics)))
        sys.stdout.flush()

        extr = cam_extrinsics[key]
        intr = cam_intrinsics[extr.camera_id]
        height = intr.height
        width = intr.width

        uid = intr.id
        R = np.transpose(qvec2rotmat(extr.qvec))
        T = np.array(extr.tvec)

        if intr.model=="SIMPLE_PINHOLE":
            focal_length_x = intr.params[0]
            FovY = focal2fov(focal_length_x, height)
            FovX = focal2fov(focal_length_x, width)
        elif intr.model=="PINHOLE":
            focal_length_x = intr.params[0]
            focal_length_y = intr.params[1]
            FovY = focal2fov(focal_length_y, height)
            FovX = focal2fov(focal_length_x, width)
        else:
            assert False, "Colmap camera model not handled: only undistorted datasets (PINHOLE or SIMPLE_PINHOLE cameras) supported!"

        image_path = os.path.join(images_folder, os.path.basename(extr.name))
        image_name = os.path.basename(image_path).split(".")[0]
        image = Image.open(image_path)

        cam_info = CameraInfo(uid=uid, R=R, T=T, FovY=FovY, FovX=FovX, image=image,
                              image_path=image_path, image_name=image_name, width=width, height=height)
        cam_infos.append(cam_info)
    sys.stdout.write('\n')
    return cam_infos
#读取points3D.ply文件，读取位置、颜色和法线，并组合成BasicPointCloud类返回
def fetchPly(path):
    plydata = PlyData.read(path)
    vertices = plydata['vertex']
    positions = np.vstack([vertices['x'], vertices['y'], vertices['z']]).T
    colors = np.vstack([vertices['red'], vertices['green'], vertices['blue']]).T / 255.0
    normals = np.vstack([vertices['nx'], vertices['ny'], vertices['nz']]).T
    return BasicPointCloud(points=positions, colors=colors, normals=normals)

def storePly(path, xyz, rgb):
    # Define the dtype for the structured array
    dtype = [('x', 'f4'), ('y', 'f4'), ('z', 'f4'),
            ('nx', 'f4'), ('ny', 'f4'), ('nz', 'f4'),
            ('red', 'u1'), ('green', 'u1'), ('blue', 'u1')]
    
    normals = np.zeros_like(xyz)

    elements = np.empty(xyz.shape[0], dtype=dtype)
    attributes = np.concatenate((xyz, normals, rgb), axis=1)
    elements[:] = list(map(tuple, attributes))

    # Create the PlyData object and write to file
    vertex_element = PlyElement.describe(elements, 'vertex')
    ply_data = PlyData([vertex_element])
    ply_data.write(path)

def readColmapSceneInfo(path, images, eval, llffhold=8):
    try:
        cameras_extrinsic_file = os.path.join(path, "sparse/0", "images.bin")
        cameras_intrinsic_file = os.path.join(path, "sparse/0", "cameras.bin")
        cam_extrinsics = read_extrinsics_binary(cameras_extrinsic_file)
        cam_intrinsics = read_intrinsics_binary(cameras_intrinsic_file)
    except:
        cameras_extrinsic_file = os.path.join(path, "sparse/0", "images.txt")
        cameras_intrinsic_file = os.path.join(path, "sparse/0", "cameras.txt")
        cam_extrinsics = read_extrinsics_text(cameras_extrinsic_file)
        cam_intrinsics = read_intrinsics_text(cameras_intrinsic_file)

    reading_dir = "images" if images == None else images
    cam_infos_unsorted = readColmapCameras(cam_extrinsics=cam_extrinsics, cam_intrinsics=cam_intrinsics, images_folder=os.path.join(path, reading_dir))
    cam_infos = sorted(cam_infos_unsorted.copy(), key = lambda x : x.image_name)
    #需要评估的话，训练集和测试集分一下，否则全部都拿来训练
    if eval:
        train_cam_infos = [c for idx, c in enumerate(cam_infos) if idx % llffhold != 0]
        test_cam_infos = [c for idx, c in enumerate(cam_infos) if idx % llffhold == 0]
    else:
        train_cam_infos = cam_infos
        test_cam_infos = []

    nerf_normalization = getNerfppNorm(train_cam_infos)

    ply_path = os.path.join(path, "sparse/0/points3D.ply")
    bin_path = os.path.join(path, "sparse/0/points3D.bin")
    txt_path = os.path.join(path, "sparse/0/points3D.txt")
    if not os.path.exists(ply_path):
        print("Converting point3d.bin to .ply, will happen only the first time you open the scene.")
        try:
            xyz, rgb, _ = read_points3D_binary(bin_path)
        except:
            xyz, rgb, _ = read_points3D_text(txt_path)
        storePly(ply_path, xyz, rgb)
    try:
        #读取ply文件
        pcd = fetchPly(ply_path)
    except:
        pcd = None
    #把这些组合成SceneInfo类返回
    scene_info = SceneInfo(point_cloud=pcd,
                           train_cameras=train_cam_infos,
                           test_cameras=test_cam_infos,
                           nerf_normalization=nerf_normalization,
                           ply_path=ply_path)
    return scene_info

def readCamerasFromTransforms(path, transformsfile, white_background, extension=".png"):
    cam_infos = []

    with open(os.path.join(path, transformsfile)) as json_file:
        contents = json.load(json_file)
        fovx = contents["camera_angle_x"]

        frames = contents["frames"]
        for idx, frame in enumerate(frames):
            cam_name = os.path.join(path, frame["file_path"] + extension)

            # NeRF 'transform_matrix' is a camera-to-world transform
            c2w = np.array(frame["transform_matrix"])
            # change from OpenGL/Blender camera axes (Y up, Z back) to COLMAP (Y down, Z forward)
            c2w[:3, 1:3] *= -1

            # get the world-to-camera transform and set R, T
            w2c = np.linalg.inv(c2w)
            R = np.transpose(w2c[:3,:3])  # R is stored transposed due to 'glm' in CUDA code
            T = w2c[:3, 3]

            image_path = os.path.join(path, cam_name)
            image_name = Path(cam_name).stem
            image = Image.open(image_path)

            im_data = np.array(image.convert("RGBA"))

            bg = np.array([1,1,1]) if white_background else np.array([0, 0, 0])

            norm_data = im_data / 255.0
            arr = norm_data[:,:,:3] * norm_data[:, :, 3:4] + bg * (1 - norm_data[:, :, 3:4])
            image = Image.fromarray(np.array(arr*255.0, dtype=np.byte), "RGB")

            fovy = focal2fov(fov2focal(fovx, image.size[0]), image.size[1])
            FovY = fovy 
            FovX = fovx

            cam_infos.append(CameraInfo(uid=idx, R=R, T=T, FovY=FovY, FovX=FovX, image=image,
                            image_path=image_path, image_name=image_name, width=image.size[0], height=image.size[1]))
            
    return cam_infos

def readNerfSyntheticInfo(path, white_background, eval, extension=".png"):
    print("Reading Training Transforms")
    train_cam_infos = readCamerasFromTransforms(path, "transforms_train.json", white_background, extension)
    print("Reading Test Transforms")
    test_cam_infos = readCamerasFromTransforms(path, "transforms_test.json", white_background, extension)
    
    if not eval:
        train_cam_infos.extend(test_cam_infos)
        test_cam_infos = []

    nerf_normalization = getNerfppNorm(train_cam_infos)

    ply_path = os.path.join(path, "points3d.ply")
    if not os.path.exists(ply_path):
        # Since this data set has no colmap data, we start with random points
        num_pts = 100_000
        print(f"Generating random point cloud ({num_pts})...")
        
        # We create random points inside the bounds of the synthetic Blender scenes
        xyz = np.random.random((num_pts, 3)) * 2.6 - 1.3
        shs = np.random.random((num_pts, 3)) / 255.0
        pcd = BasicPointCloud(points=xyz, colors=SH2RGB(shs), normals=np.zeros((num_pts, 3)))

        storePly(ply_path, xyz, SH2RGB(shs) * 255)
    try:
        pcd = fetchPly(ply_path)
    except:
        pcd = None

    scene_info = SceneInfo(point_cloud=pcd,
                           train_cameras=train_cam_infos,
                           test_cameras=test_cam_infos,
                           nerf_normalization=nerf_normalization,
                           ply_path=ply_path)
    return scene_info
#字典，存储对应的回调函数
sceneLoadTypeCallbacks = {
    "Colmap": readColmapSceneInfo,
    "Blender" : readNerfSyntheticInfo
}