# # 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 import torch.nn.functional as F from random import randint from utils.loss_utils import l1_loss, ssim, or_loss from utils.general_utils import get_expon_lr_func from gaussian_renderer import render, network_gui import sys from scene import Scene, GaussianModel from utils.general_utils import safe_state import uuid from tqdm import tqdm from utils.image_utils import psnr, vis_orient from argparse import ArgumentParser, Namespace from arguments import ModelParams, PipelineParams, OptimizationParams import pickle as pkl 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 tb_writer = prepare_output_and_logger(dataset) gaussians = GaussianModel(dataset.sh_degree) scene = Scene(dataset, gaussians) gaussians.training_setup(opt) if checkpoint: (model_params, first_iter) = torch.load(checkpoint) gaussians.restore(model_params, opt) if dataset.trainable_cameras or dataset.trainable_intrinsics: params_cam_rotation = {} params_cam_translation = {} params_cam_fov = {} for k in scene.train_cameras.keys(): for camera in scene.train_cameras[k]: if dataset.trainable_cameras: params_cam_rotation[camera.image_name] = camera._rotation_res params_cam_translation[camera.image_name] = camera._translation_res if dataset.trainable_intrinsics: params_cam_fov[camera.image_name] = camera._fov_res params_cam = list(params_cam_rotation.values()) + list(params_cam_translation.values()) + list(params_cam_fov.values()) l = [ {'params': list(params_cam_rotation.values()), 'lr': opt.cam_rotation_lr, "name": "rotation"}, {'params': list(params_cam_translation.values()), 'lr': opt.cam_translation_lr_init * gaussians.spatial_lr_scale, "name": "translation"}, {'params': list(params_cam_fov.values()), 'lr': opt.cam_fov_lr, "name": "fov"} ] optimizer_cameras = torch.optim.Adam(l, lr=0.0, eps=1e-15) translation_scheduler_args = get_expon_lr_func(lr_init=opt.cam_translation_lr_init * gaussians.spatial_lr_scale, lr_final=opt.cam_translation_lr_final * gaussians.spatial_lr_scale, max_steps=opt.cam_lr_max_steps) bg_color = [1, 1, 1, 0, 0, 0, 0, 0, 0, 100] if dataset.white_background else [0, 0, 0, 0, 0, 0, 0, 0, 0, 100] 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="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 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() viewpoint_cam = viewpoint_stack.pop(randint(0, len(viewpoint_stack)-1)) # Render if (iteration - 1) == debug_from: pipe.debug = True render_pkg = render(viewpoint_cam, gaussians, pipe, background) image = render_pkg["render"] mask = render_pkg["mask"] orient_angle = render_pkg["orient_angle"] orient_conf = render_pkg["orient_conf"] viewspace_point_tensor = render_pkg["viewspace_points"] visibility_filter = render_pkg["visibility_filter"] radii = render_pkg["radii"] # Loss gt_image = viewpoint_cam.original_image.cuda() gt_mask = viewpoint_cam.original_mask.cuda() gt_orient_angle = viewpoint_cam.original_orient_angle.cuda() gt_orient_conf = viewpoint_cam.original_orient_conf.cuda() Ll1 = l1_loss(image, gt_image, mask=gt_mask[1:].detach()) Lssim = (1.0 - ssim(image * gt_mask[1:], gt_image * gt_mask[1:])) Lmask = l1_loss(mask, gt_mask) orient_weight = torch.ones_like(gt_mask[:1]) * gt_orient_conf Lorient = or_loss(orient_angle, gt_orient_angle, orient_conf, weight=orient_weight, mask=gt_mask[:1]) if torch.isnan(Lorient).any(): Lorient = torch.zeros_like(Ll1) loss = ( Ll1 * opt.lambda_dl1 + Lssim * opt.lambda_dssim + Lmask * opt.lambda_dmask + Lorient * opt.lambda_dorient ) 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, Lmask, Lorient, loss, l1_loss, or_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: # 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: for param in [gaussians._xyz, gaussians._features_dc, gaussians._features_rest, gaussians._opacity, gaussians._label, gaussians._scaling, gaussians._rotation]: if param.grad is not None and param.grad.isnan().any(): gaussians.optimizer.zero_grad(set_to_none = True) print('NaN during backprop was found, skipping iteration...') gaussians.optimizer.step() gaussians.optimizer.zero_grad(set_to_none = True) if iteration < opt.iterations_cam and dataset.trainable_cameras: ''' Learning rate scheduling per step ''' for param_group in optimizer_cameras.param_groups: if param_group["name"] == "translation": lr = translation_scheduler_args(iteration) param_group['lr'] = lr for param in params_cam: if param.grad is not None and param.grad.isnan().any(): optimizer_cameras.zero_grad(set_to_none = True) print('NaN during backprop was found, skipping iteration...') optimizer_cameras.step() optimizer_cameras.zero_grad() if (iteration in checkpoint_iterations): print("\n[ITER {}] Saving Checkpoint".format(iteration)) os.makedirs(scene.model_path + "/checkpoints", exist_ok=True) os.makedirs(scene.model_path + "/cameras", exist_ok=True) torch.save((gaussians.capture(), iteration), scene.model_path + "/checkpoints/" + str(iteration) + ".pth") if dataset.trainable_cameras: pkl.dump((params_cam_rotation, params_cam_translation, params_cam_fov), open(scene.model_path + "/cameras/" + str(iteration) + ".pkl", 'wb')) projection_all = {} for camera in scene.train_cameras[1.0]: projection_all[camera.image_name] = camera.full_proj_transform.cpu() pkl.dump(projection_all, open(scene.model_path + "/cameras/" + str(iteration) + "_matrices.pkl", 'wb')) 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]) # 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, Lmask, Lorient, loss, l1_loss, or_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/ce_loss', Lmask.item(), iteration) tb_writer.add_scalar('train_loss_patches/or_loss', Lorient.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: Ll1_test = 0.0 Lmask_test = 0.0 Lorient_test = 0.0 psnr_test = 0.0 for idx, viewpoint in enumerate(config['cameras']): render_pkg = renderFunc(viewpoint, scene.gaussians, *renderArgs) image = torch.clamp(render_pkg["render"], 0.0, 1.0) mask = torch.clamp(render_pkg["mask"], 0.0, 1.0) orient_angle = torch.clamp(render_pkg["orient_angle"], 0.0, 1.0) orient_conf = render_pkg["orient_conf"] orient_conf_vis = (1 - 1 / (orient_conf + 1)) * mask[:1] gt_image = torch.clamp(viewpoint.original_image.to("cuda"), 0.0, 1.0) gt_mask = torch.clamp(viewpoint.original_mask.to("cuda"), 0.0, 1.0) gt_orient_angle = torch.clamp(viewpoint.original_orient_angle.to("cuda"), 0.0, 1.0) gt_orient_conf = viewpoint.original_orient_conf.to("cuda") gt_orient_conf_vis = (1 - 1 / (gt_orient_conf + 1)) * gt_mask[:1] if tb_writer and (idx < 5): tb_writer.add_images(config['name'] + "_view_{}/render".format(viewpoint.image_name), image[None], global_step=iteration) tb_writer.add_images(config['name'] + "_view_{}/render_mask".format(viewpoint.image_name), F.pad(mask, (0, 0, 0, 0, 0, 3-mask.shape[0]), 'constant', 0)[None], global_step=iteration) tb_writer.add_images(config['name'] + "_view_{}/render_orient".format(viewpoint.image_name), vis_orient(orient_angle, mask[:1])[None], global_step=iteration) tb_writer.add_images(config['name'] + "_view_{}/render_orient_conf".format(viewpoint.image_name), vis_orient(orient_angle, orient_conf_vis)[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) tb_writer.add_images(config['name'] + "_view_{}/ground_truth_mask".format(viewpoint.image_name), F.pad(gt_mask, (0, 0, 0, 0, 0, 3-gt_mask.shape[0]), 'constant', 0)[None], global_step=iteration) tb_writer.add_images(config['name'] + "_view_{}/ground_truth_orient".format(viewpoint.image_name), vis_orient(gt_orient_angle, gt_mask[:1])[None], global_step=iteration) tb_writer.add_images(config['name'] + "_view_{}/ground_truth_orient_conf".format(viewpoint.image_name), vis_orient(gt_orient_angle, gt_orient_conf_vis)[None], global_step=iteration) Ll1_test += l1_loss(image, gt_image).mean().double() Lmask_test += l1_loss(mask, gt_mask).mean().double() Lorient_test += or_loss(orient_angle, gt_orient_angle, mask=gt_mask[:1], weight=gt_orient_conf).mean().double() psnr_test += psnr(image, gt_image).mean().double() Ll1_test /= len(config['cameras']) Lmask_test /= len(config['cameras']) Lorient_test /= len(config['cameras']) psnr_test /= len(config['cameras']) print("\n[ITER {}] Evaluating {}: L1 {} CE {} OR {} PSNR {}".format(iteration, config['name'], Ll1_test, Lmask_test, Lorient_test, psnr_test)) if tb_writer: tb_writer.add_scalar(config['name'] + '/loss_viewpoint - l1_loss', Ll1_test, iteration) tb_writer.add_scalar(config['name'] + '/loss_viewpoint - ce_loss', Lmask_test, iteration) tb_writer.add_scalar(config['name'] + '/loss_viewpoint - or_loss', Lorient_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_histogram("scene/label_histogram", scene.gaussians.get_label, 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=[1_000, 5_000, 15_000, 30_000]) parser.add_argument("--save_iterations", nargs="+", type=int, default=[1_000, 5_000, 15_000, 30_000]) parser.add_argument("--quiet", action="store_true") parser.add_argument("--checkpoint_iterations", nargs="+", type=int, default=[1_000, 5_000, 15_000, 30_000]) 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) 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) 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.")