train_vast.py

#
# Copyright (C) 2023, Inria
# GRAPHDECO research group, https://team.inria.fr/graphdeco
# All rights reserved.
#
# This software is free for non-commercial, research and evaluation use 
# under the terms of the LICENSE.md file.
#
# For inquiries contact  george.drettakis@inria.fr
#

import logging
import os
import copy
from glob import glob
import torch
from random import randint
from utils.loss_utils import l1_loss, ssim
from gaussian_renderer import render, network_gui
import sys
from scene import Scene, GaussianModel, PartitionScene
from scene.vastgs.appearance_network import decouple_appearance
from utils.general_utils import safe_state
from utils.partition_utils import data_partition, read_camList
import uuid
from tqdm import tqdm
from utils.image_utils import psnr
from utils.manhattan_utils import get_man_trans
from argparse import ArgumentParser, Namespace
from arguments import ModelParams, PipelineParams, OptimizationParams
import multiprocessing as mp
from seamless_merging import seamless_merge


try:
    from torch.utils.tensorboard import SummaryWriter
    TENSORBOARD_FOUND = True
except ImportError:
    TENSORBOARD_FOUND = False


def training(dataset, opt, pipe, testing_iterations, saving_iterations, checkpoint_iterations, checkpoint, debug_from, logger=None):
    # read train and test camera list
    test_camList = read_camList(dataset.model_path + "/test_cameras.txt")

    first_iter = 0
    tb_writer = prepare_output_and_logger(dataset)
    gaussians = GaussianModel(dataset.sh_degree)
    # scene = Scene(dataset, gaussians)
    scene = PartitionScene(dataset, gaussians)
    gaussians.training_setup(opt)
    if checkpoint:
        (model_params, first_iter) = torch.load(checkpoint)
        gaussians.restore(model_params, opt)

    bg_color = [1, 1, 1] if dataset.white_background else [0, 0, 0]
    background = torch.tensor(bg_color, dtype=torch.float32, device="cuda")

    iter_start = torch.cuda.Event(enable_timing = True)
    iter_end = torch.cuda.Event(enable_timing = True)

    viewpoint_stack = None
    ema_loss_for_log = 0.0
    progress_bar = tqdm(range(first_iter, opt.iterations), desc=f"Training progress Partition: {dataset.partition_id}")
    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

        iter_start.record()

        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
        if not viewpoint_stack:
            viewpoint_stack = scene.getTrainCameras().copy()
            new_viewpoint_stack = []
            for view in viewpoint_stack:  # 训练时剔除测试集图片
                if view.image_name not in test_camList:
                    new_viewpoint_stack.append(view)
            viewpoint_stack = new_viewpoint_stack
        viewpoint_cam = viewpoint_stack.pop(randint(0, len(viewpoint_stack)-1))

        # Render
        if (iteration - 1) == debug_from:
            pipe.debug = True

        bg = torch.rand((3), device="cuda") if opt.random_background else background

        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"]
        # decouple appearance model
        decouple_image, transformation_map = decouple_appearance(image, gaussians, viewpoint_cam.uid)
        gt_image = viewpoint_cam.original_image.cuda()

        # if viewpoint_cam.image_name in test_camList:
            # # 如果该图片在测试集中，移除该图像的右半边用于test，仅使用左半边图像进行train
            # gt_image = gt_image[..., :gt_image.shape[-1] // 2]
            # image = image[..., :image.shape[-1] // 2]
            # decouple_image = decouple_image[..., :decouple_image.shape[-1] // 2]

        # Loss
        # Ll1 = l1_loss(image, gt_image)
        Ll1 = l1_loss(decouple_image, gt_image)
        loss = (1.0 - opt.lambda_dssim) * Ll1 + opt.lambda_dssim * (1.0 - ssim(image, gt_image))
        loss.backward()

        iter_end.record()

        with torch.no_grad():
            # Progress bar
            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), logger=logger)
            if (iteration in saving_iterations):
                if logger is not None:
                    logger.info(f"Saving Gaussians at iteration {iteration}")
                print("\n[ITER {}] Saving Gaussians".format(iteration))
                scene.save(iteration)

            # Densification
            if iteration < opt.densify_until_iter:
                # 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])
                gaussians.add_densification_stats(viewspace_point_tensor, visibility_filter)

                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)
                
                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:
                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")


def parallel_local_training(gpu_id, partition_id, lp_args, op_args, pp_args, test_iterations, save_iterations, checkpoint_iterations,
                            start_checkpoint, debug_from):
    torch.cuda.set_device(gpu_id)

    partition_model_path = f"{lp_args.model_path}/partition_point_cloud/visible"
    lp_args.partition_id = partition_id
    lp_args.partition_model_path = partition_model_path

    logger = setup_logging(partition_id, file_path=partition_model_path)
    # 启动训练
    logger.info("Starting process")
    training(lp_args, op_args, pp_args, test_iterations, save_iterations, checkpoint_iterations,start_checkpoint, debug_from, logger=logger)
    logger.info("Finishing process")


def setup_logging(partition_id, file_path):
    # 创建一个 logger
    logger = logging.getLogger(f'Client_{partition_id}')
    logger.setLevel(logging.INFO)  # 设置日志级别

    # 创建文件 handler，用于写入日志文件
    if not os.path.exists(file_path):
        os.makedirs(file_path)
    file_handler = logging.FileHandler(f'{file_path}/Partition_{partition_id}.log')

    # 创建 formatter
    formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
    file_handler.setFormatter(formatter)

    # 添加 handler 到 logger
    logger.addHandler(file_handler)

    return logger


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())
    #     args.model_path = os.path.join("./output/", unique_str[0:10])

    if not args.model_path:
        model_path = os.path.join("./output/", args.exp_name)
        # 如果这个文件存在，就在这个文件名的基础上创建新的文件夹，文件名后面跟上1,2,3
        if os.path.exists(model_path):
            base_name = os.path.basename(model_path)
            dir_name = os.path.dirname(model_path)
            file_name, file_ext = os.path.splitext(base_name)
            counter = 1
            while os.path.exists(os.path.join(dir_name, f"{file_name}_{counter}{file_ext}")):
                counter += 1
            new_folder_name = f"{file_name}_{counter}{file_ext}"
            model_path = os.path.join(dir_name, new_folder_name)
        args.model_path = model_path
        
    # Set up output folder
    print("Output folder: {}".format(args.model_path))
    os.makedirs(args.model_path, exist_ok=True)
    with open(os.path.join(args.model_path, "cfg_args"), 'w') as cfg_log_f:
        var_dict = copy.deepcopy(vars(args))
        del_var_list = ["manhattan", "man_trans", "pos", "rot",
                        "m_region", "n_region", "extend_rate", "visible_rate",
                        "num_gpus", "partition_id", "partition_model_path", "platform",
                        "llffhold"]  # 删除多余的变量，防止无法使用SIBR可视化
        for key in vars(args).keys():
            if key in del_var_list:
                del var_dict[key]
        cfg_log_f.write(str(Namespace(**var_dict)))

    # 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

def training_report(tb_writer, iteration, Ll1, loss, l1_loss, elapsed, testing_iterations, scene : Scene, renderFunc, renderArgs, logger=None):
    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'])
                if logger is not None:
                    logger.info("[ITER {}] Evaluating {}: L1 {} PSNR {}".format(iteration, config['name'], l1_test, psnr_test))
                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, 60_000])
    parser.add_argument("--save_iterations", nargs="+", type=int, default=[7_000, 30_000, 60_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)

    lp, op, pp = lp.extract(args), op.extract(args), pp.extract(args)

    # Initialize system state (RNG)
    safe_state(args.quiet)

    # Start GUI server, configure and run training
    network_gui.init(args.ip, args.port)
    torch.autograd.set_detect_anomaly(args.detect_anomaly)

    # Manhattan Alignment
    lp.man_trans = get_man_trans(lp)

    # train multi gpu
    mp.set_start_method('spawn', force=True)
    tb_writer = prepare_output_and_logger(lp)

    # data partition
    partition_num, partition_id_list = data_partition(lp)

    cuda_devices = torch.cuda.device_count()
    print(f"Found {cuda_devices} CUDA devices")
    training_round = partition_num // cuda_devices
    remainder = partition_num % cuda_devices

    # Main Loops
    for i in range(training_round):
        partition_pool = [i + training_round * j for j in range(cuda_devices)]

        processes = []
        for index, device_id in enumerate(range(cuda_devices)):
            partition_index = partition_pool[index]
            partition_id = partition_id_list[partition_index]
            print("train partition {} on gpu {}".format(partition_id, device_id))
            p = mp.Process(target=parallel_local_training, name=f"Partition_{partition_id}",
                        args=(device_id, partition_id, lp, op, pp,
                              args.test_iterations, args.save_iterations, args.checkpoint_iterations,
                              args.start_checkpoint, args.debug_from))
            processes.append(p)
            p.start()

        for p in processes:
            p.join()  # 等待所有进程完成
            # processes = []

        torch.cuda.empty_cache()

    if remainder != 0:
        partition_pool = [cuda_devices * training_round + i for i in range(remainder)]
        processes = []
        for index, device_id in enumerate(range(cuda_devices)[:remainder]):
            # torch.cuda.set_device(device_id)
            partition_index = partition_pool[index]
            partition_id = partition_id_list[partition_index]
            print("train partition {} on gpu {}".format(partition_id, device_id))
            p = mp.Process(target=parallel_local_training, name=f"Partition_{partition_id}",
                        args=(device_id, partition_id, lp, op, pp,
                              args.test_iterations, args.save_iterations, args.checkpoint_iterations,
                              args.start_checkpoint, args.debug_from))
            processes.append(p)
            p.start()

        for p in processes:
            p.join()

        torch.cuda.empty_cache()

    print("\nTraining complete.")

    # seamless_merging 无缝合并
    print("Merging Partitions...")
    all_point_cloud_dir = glob(os.path.join(lp.model_path, "point_cloud", "*"))

    for point_cloud_dir in all_point_cloud_dir:
        seamless_merge(lp.model_path, point_cloud_dir)

    # All done
    print("All Done!")