train.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 os
import torch
from random import randint
from utils.loss_utils import l1_loss, ssim, cos_loss
from gaussian_renderer import render
import sys
from scene import Scene, GaussianModel

import uuid
from tqdm import tqdm
from utils.image_utils import psnr,  depth2normal, normal2curv
from argparse import ArgumentParser, Namespace
import os
os.environ['CUDA_LAUNCH_BLOCKING'] = '1'
from arguments import ModelParams, PipelineParams, OptimizationParams
from utils.system_utils import set_seed
import utils.polarization_utils as polar_utils
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):
    first_iter = 0
    save_current_files(dataset)
    tb_writer = prepare_output_and_logger(dataset)


    gaussians = GaussianModel(dataset)
    scene = Scene(dataset, gaussians, opt.camera_lr, shuffle=False, resolution_scales=[1])
    use_mask = dataset.use_mask
    gaussians.training_setup(opt)
    if checkpoint:
        (model_params, first_iter) = torch.load(checkpoint)
        gaussians.restore(model_params, opt)
    elif use_mask: # visual hull init
        gaussians.mask_prune(scene.getTrainCameras(), 4)
        None

    opt.densification_interval = max(opt.densification_interval, len(scene.getTrainCameras()))

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

    iter_start = torch.cuda.Event(enable_timing = True)
    iter_end = torch.cuda.Event(enable_timing = True)
    pool = torch.nn.MaxPool2d(9, stride=1, padding=4)

    viewpoint_stack = None
    ema_loss_for_log = 0.0
    progress_bar = tqdm(range(first_iter, opt.iterations), desc="Training progress")
    first_iter += 1
    initial_stage = True
    viewpoint_stack = scene.getTrainCameras(1).copy()[:]
    view_num = len(viewpoint_stack)
    depth_list = torch.zeros_like(viewpoint_stack[0].mono[3:]).repeat(view_num,1,1)
    for iteration in range(first_iter, opt.iterations + 2):
        loss = 0
        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()

        scale = 1
        if iteration> opt.initial_iterations:
            initial_stage = False
        # Pick a random Camera
        
        if not viewpoint_stack:
            viewpoint_stack = scene.getTrainCameras(scale).copy()[:]
        viewpoint_cam= viewpoint_stack.pop(randint(0, len(viewpoint_stack) - 1))

        # Render
        if (iteration - 1) == debug_from:
            pipe.debug = True
        
        background = torch.rand((3), dtype=torch.float32, device="cuda") if dataset.random_background else background
        patch_size = [float('inf'), float('inf')]


        render_pkg = render(viewpoint_cam, gaussians, pipe, background, patch_size, initial_stage=initial_stage)
        image, normal, depth, opac, viewspace_point_tensor, visibility_filter, radii = \
            render_pkg["render"], render_pkg["normal"], render_pkg["depth"], render_pkg["opac"], \
            render_pkg["viewspace_points"], render_pkg["visibility_filter"], render_pkg["radii"]
        if not initial_stage:
            image = render_pkg['rendered_image_with_ref']
            rendered_normal_world = render_pkg['rendered_normal_world']

        mask_gt = viewpoint_cam.get_gtMask(use_mask)
        gt_image = viewpoint_cam.get_gtImage(background, use_mask)
        if viewpoint_cam.s0 is not None:
            gt_image, gt_s1, gt_s2 = viewpoint_cam.get_gtStokes(background, use_mask)
        mask_vis = (opac.detach() > 1e-5)
        normal = torch.nn.functional.normalize(normal, dim=0) * mask_vis
        d2n = depth2normal(depth, mask_vis, viewpoint_cam)
        ##########
        depth_list[viewpoint_cam.uid] = depth.detach()
        loss_azimuth = 0
        if iteration > opt.azimuth_start:
            bs = depth.permute(1,2,0).view(-1,1)
            idx = torch.randint(bs.shape[0], [2048])
            azimuth_info = polar_utils.mvas_net(
                depth.permute(1,2,0).view(-1,1)[idx], # [512,1]
                viewpoint_cam.view_direction.view(-1,3)[idx],  # [512,3]
                viewpoint_cam.camera_center, # [3]
                mask_gt.permute(1,2,0).view(-1,1)[idx], # [512,1]
                device="cuda", 
                dataset=scene.dataset_all, 
                depth_list=depth_list,
                threshold=opt.threshold,
                training=True)

            loss_azimuth = polar_utils.get_azimuth_loss(rendered_normal_world.view(-1,3)[idx], azimuth_info)
        #################

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

        if viewpoint_cam.s0 is not None and initial_stage is False:
            s1 = render_pkg['rendered_s1']
            s2 = render_pkg['rendered_s2']
            Ls1 = l1_loss(s1, gt_s1)
            Ls2 = l1_loss(s2, gt_s2)
            loss_stokes = Ls1+Ls2
            loss += opt.stokes_l*loss_stokes
        else:
            loss_stokes = 0

        loss_mask = (opac * (1 - pool(mask_gt))).mean()

        loss_surface = cos_loss(normal, d2n)

        opac_ = gaussians.get_opacity
        opac_mask0 = torch.gt(opac_, 0.01) * torch.le(opac_, 0.2)
        opac_mask1 = torch.gt(opac_, 0.8) * torch.le(opac_, 0.99)
        opac_mask2 = torch.gt(opac_, 0.2) * torch.le(opac_, 0.8)
        opac_mask = opac_mask0 * 0.01 + opac_mask1 * 0.01 + opac_mask2 
        loss_opac = (torch.exp(-(opac_ - 0.5)**2 * 20) * opac_mask).mean()

        curv_n = normal2curv(normal, mask_vis)
        loss_curv = l1_loss(curv_n * 1, 0) #+ 1 * l1_loss(curv_d2n, 0)
        loss += 1 * loss_rgb
        loss += 0.1 * loss_mask
        loss += (0.01 + 0.1 * iteration / opt.iterations) * loss_surface * 5
        loss += 0.005 * loss_curv
        loss += 0.01* loss_opac

        
        if opt.azimuth_l > 0 and iteration > opt.azimuth_start:
            loss += opt.azimuth_l * loss_azimuth

        loss.backward()
        iter_end.record()

        with torch.no_grad():
            # Progress bar
            ema_loss_for_log = 0.4 * loss_rgb.item() + 0.6 * ema_loss_for_log
            if iteration % 10 == 0:
                progress_bar.set_postfix({"Loss": f"{ema_loss_for_log:.{7}f}, Pts={len(gaussians._xyz)}"})
                progress_bar.update(10)
            if iteration == opt.iterations:
                progress_bar.close()

            # Log and save
            test_background = torch.tensor([1, 1, 1] if dataset.white_background else [0, 0, 0], dtype=torch.float32, device="cuda")
            training_report(tb_writer, iteration, Ll1, loss, l1_loss, iter_start.elapsed_time(iter_end), 
                            testing_iterations, scene, pipe, test_background, use_mask, loss_stokes=loss_stokes,loss_azimuth=loss_azimuth)
            if (iteration in saving_iterations):
                print("\n[ITER {}] Saving Gaussians".format(iteration))
                scene.save(iteration)

            # Densification
            if iteration > opt.densify_from_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)
                min_opac = 0.1
                if iteration % opt.densification_interval == 0:
                    gaussians.adaptive_prune(min_opac, scene.cameras_extent)
                    gaussians.adaptive_densify(opt.densify_grad_threshold, scene.cameras_extent)
                
                if (iteration - 1) % opt.opacity_reset_interval == 0 and opt.opacity_lr > 0:
                    gaussians.reset_opacity(0.12, iteration)


            if iteration < opt.iterations:
                gaussians.optimizer.step()
                gaussians.optimizer.zero_grad()

            if (iteration in checkpoint_iterations):
                # gaussians.adaptive_prune(min_opac, scene.cameras_extent)
                print("\n[ITER {}] Saving Checkpoint".format(iteration))
                torch.save((gaussians.capture(), iteration), scene.model_path + "/chkpnt" + str(iteration) + ".pth")


def save_current_files(args):
    print("Saving current code files to output folder for reproducibility.")
    import shutil
    save_dir = os.path.join(args.model_path, "code")
    os.makedirs(save_dir, exist_ok = True)
    current_path = os.getcwd()
    shutil.copytree(os.path.join(current_path, "gaussian_renderer"), os.path.join(save_dir, "gaussian_renderer"))
    shutil.copy2(os.path.join(current_path, "train.py"), os.path.join(save_dir, "train.py"))
    shutil.copytree(os.path.join(current_path, "arguments"), os.path.join(save_dir, "arguments"))
    shutil.copytree(os.path.join(current_path, "scene"), os.path.join(save_dir, "scene"))
    shutil.copytree(os.path.join(current_path, "utils"), os.path.join(save_dir, "utils"))
    for file in os.listdir(current_path):
        if file.endswith(".py"):
            shutil.copy2(os.path.join(current_path, file), os.path.join(save_dir, file))
        if file.endswith(".md"):
            shutil.copy2(os.path.join(current_path, file), os.path.join(save_dir, file))
        if file.endswith(".bash"):
            shutil.copy2(os.path.join(current_path, file), os.path.join(save_dir, file))
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", f"{args.source_path.split('/')[-1]}_{unique_str[0:10]}")
        
        
    # 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:
        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

def training_report(tb_writer, iteration, Ll1, loss, l1_loss, elapsed, testing_iterations, scene : Scene, pipe, bg, use_mask, loss_azimuth = 0, loss_stokes = 0):
    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('train_loss_patches/loss_azimuth', loss_azimuth, iteration)
        tb_writer.add_scalar('train_loss_patches/loss_stokes', loss_stokes, 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()[::8]})
        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(render(viewpoint, scene.gaussians, pipe, bg, [float('inf'), float('inf')],initial_stage=False)["render"], 0.0, 1.0)
                    gt_image = torch.clamp(viewpoint.get_gtImage(bg, with_mask=use_mask), 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
    # set_seed(2463)
    set_seed(953)
    parser = ArgumentParser(description="Training script parameters")
    lp = ModelParams(parser)
    op = OptimizationParams(parser)
    pp = PipelineParams(parser)
    torch.set_default_tensor_type('torch.cuda.FloatTensor')

    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=[5_000, 10_000, 15_000, 20_000, 25_000, 30_000])
    parser.add_argument("--save_iterations", nargs="+", type=int, default=[5_000, 10_000, 15_000, 20_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)
    args.eval = False
    # args.iterations = 15000
    print("Optimizing " + args.model_path)
    
    torch.autograd.set_detect_anomaly(args.detect_anomaly)
    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)

    # All done
    print("\nTraining complete.")