# # 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 time 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, eghuman_dataset, collate_fn, Eghuman from src.deform_model import Deform_Model from scene.gaussian_nograd_model import GaussianModel_nograd from scene.test_data import TestData from utils.general_utils import safe_state import uuid from tqdm import tqdm import numpy as np from utils.image_utils import psnr from argparse import ArgumentParser, Namespace from arguments import ModelParams, PipelineParams, OptimizationParams from utils.sh_utils import RGB2SH from torch.utils.data import DataLoader import torch.nn.functional as F import torchvision.models as models import torch.nn as nn import torchvision from simple_knn._C import distCUDA2 # from torch_geometric.data import Data # from torch_geometric.transforms import SamplePoints try: from torch.utils.tensorboard import SummaryWriter TENSORBOARD_FOUND = False except ImportError: TENSORBOARD_FOUND = False os.environ["CUDA_VISIBLE_DEVICES"] = "7" os.environ['CUDA_LAUNCH_BLOCKING'] = '1' def encoding_position(xyz_sample): L = 10 posison_encoding = [] for i in range(L): posison_encoding.append(torch.sin(2**i*np.pi*xyz_sample)) posison_encoding.append(torch.cos(2**i*np.pi*xyz_sample)) posison_encoding.append(torch.sin(2**L*np.pi*xyz_sample)) posison_encoding = torch.concat(posison_encoding, -1) return posison_encoding def requires_grad(model, flag=True): for p in model.parameters(): p.requires_grad = flag def gs_requires_grad(gs, flag): for i in gs: for p in i['params']: p.requires_grad = flag # gs._xyz_scale = gs._xyz_scale.detach() # gs._align_R = gs._align_R.detach() # gs._align_T = gs._align_T.detach() # gs._features_dc = gs._features_dc.detach() # gs._features_rest = gs._features_rest.detach() # gs._opacity = gs._opacity.detach() # gs._scaling = gs._scaling.detach() # gs ._rotation = gs._rotation.detach() def training(train_loader, dataset, opt, pipe, testing_iterations, saving_iterations, checkpoint_iterations, checkpoint, debug_from): DeformModel = Deform_Model(dataset.data_device).to(dataset.data_device) eghuman = Eghuman(opt, dataset.data_device, dataset.use_sh, dataset.sh_degree, dataset.checkpoint_path, dataset.deca_pretain_path, DeformModel.deformNet) data_tool = TestData(dataset.data_device) eghuman.feature_net.eval() eghuman.gs_decoder.eval() # eghuman.pose_flame.eval() eghuman.deformNet.eval() start_iter = eghuman.set_up() iter = 0 if dataset.checkpoint_path is not None: iter = start_iter + 1 for epoch in range(opt.epoch): times = [] for step ,data_train in enumerate(train_loader): gaussians = data_train[0]['gs'] scene = data_train[0]['sc'] dataset = data_train[0]['args'] data_length = data_train[0]['data_length'] first_iter = 0 tb_writer = prepare_output_and_logger(dataset) 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=dataset.data_device) viewpoint_stack = None first_iter += 1 # preframe_index = [i for i in range(30, len(scene.getTrainCameras().copy()), 30)] preframe_index = [2,3,13] viewpoint_stack = scene.getTrainCameras().copy() input_img = [] input_transform = [] for frame_iter in range(len(preframe_index)): input_viewpoint = viewpoint_stack[preframe_index[frame_iter]] input_transform.append(input_viewpoint.full_proj_transform) input_img.append(input_viewpoint.original_image.to(dataset.data_device)) input_img = torch.stack(input_img, 0) flame_input = data_tool.image_crop(input_img[1, ...].permute(1, 2, 0).cpu().numpy()) 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 # init flame eghuman.opt.zero_grad() start_time = time.time() parameters = eghuman.shape_flame(flame_input.to('cuda:7')) codedict_shape = data_tool.decompose_code(parameters) # codedict_pose = eghuman.pose_flame(flame_input, opt.n_eyes, opt.n_jaw, opt.n_neck, opt.n_person) # codedict = {**codedict_shape, **codedict_pose} codedict = codedict_shape codedict['trans_params'] = eghuman.trans_params DeformModel.example_init(codedict) # Pick a person time sequeece if not viewpoint_stack: viewpoint_stack = scene.getTrainCameras().copy() gt_viewpoint = viewpoint_stack.pop(randint(0, len(viewpoint_stack)-1)) # deform gaussians verts_final, rot_delta, scale_coef, verts_flame = DeformModel.decode(codedict) end_time = time.time() execution_time = end_time - start_time times.append(execution_time) # print(f"代码段执行时间为: {execution_time} 秒") if iteration == 1: gaussians.create_from_verts(verts_flame[0]) gaussians.update_xyz_rot_scale(verts_flame[0], rot_delta[0], scale_coef[0]) else: gaussians.update_xyz_rot_scale(verts_final[0], rot_delta[0], scale_coef[0]) # decoder gs parameters point_uv, point_z = eghuman.uv_compute(gaussians.get_xyz, input_transform, input_img, dataset.data_device) z_embedding = encoding_position(point_z) feature_map = eghuman.feature_net(input_img.to('cuda:7')) point_fc2d = eghuman.uv_sample(input_img, feature_map, point_uv) gs_para = eghuman.gs_decoder(gaussians.get_xyz, point_fc2d, z_embedding) gaussians._features_dc = gs_para['f_dc'] gaussians._opacity = gs_para['opacity'] gaussians._scaling = gs_para['scaling'] gaussians._rotation = gs_para['rotation'] if dataset.use_sh: render_pkg = render(gt_viewpoint, gaussians, pipe, background) else: render_pkg = render(gt_viewpoint, gaussians, pipe, background, override_color = gaussians._features_dc) image_parameter, _, _, _ = render_pkg["render"], render_pkg["viewspace_points"], render_pkg["visibility_filter"], render_pkg["radii"] torchvision.utils.save_image(image_parameter, os.path.join('/data_160TB/2021/yangkun/gha-gaussian/train_imgs', '{0:05d}'.format(iteration) + "_rgb.png")) # Loss gt_image = (gt_viewpoint.original_image).to(dataset.data_device) torchvision.utils.save_image(gt_image, os.path.join('/data_160TB/2021/yangkun/gha-gaussian/train_imgs', '{0:05d}'.format(iteration) + "_gt.png")) # with open('/data_160TB/2021/yangkun/gha-gaussian/checkpoints/test/3_view_input/checkpoint_allviews_data_new_2loss_1104_x16/time_log.txt', 'a') as file: # file.write(f'time: {execution_time}\n') if (iteration in saving_iterations): #save print("\n[ITER {}] Saving FLAME".format(iteration)) scene.save(iteration, 'ply') if dataset.type_3d == 'obj': scene.save(iteration, dataset.type_3d) if (iteration in checkpoint_iterations): print("\n[ITER {}] Saving Checkpoint".format(iteration)) torch.save((gaussians.capture(), iteration), scene.model_path + "/chkpnt" + str(iteration) + ".pth") times_avg = sum(times) / len(times) # with open('/data_160TB/2021/yangkun/gha-gaussian/checkpoints/test/3_view_input/checkpoint_allviews_data_new_2loss_1104_x16/time_log.txt', 'a') as file: # file.write(f'time_avg: {times_avg}\n') print(times_avg) 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, 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:7"), 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="127.0.0.2") parser.add_argument('--port', type=int, default=6061) 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=[]) parser.add_argument("--save_iterations", nargs="+", type=int, default=[1, 200, 500, 1_000, 2_000, 3_000, 4_000, 5_000, 6_000, 7_000, 8_000, 14_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) 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) dataset_train = eghuman_dataset(lp.extract(args), train = True) dataloader_train = DataLoader(dataset=dataset_train, batch_size=1, shuffle=False, num_workers=0, collate_fn=collate_fn) training(dataloader_train, 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.")