# # 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 torch import torch.nn.functional as F from scene import Scene import os from tqdm import tqdm from os import makedirs from gaussian_renderer import render import torchvision from utils.general_utils import safe_state from utils.image_utils import vis_orient from utils.graphics_utils import fov2focal from argparse import ArgumentParser from arguments import ModelParams, PipelineParams, OptimizationParams, get_combined_args from gaussian_renderer import GaussianModel import math import pickle as pkl def render_set(model_path, name, iteration, views, gaussians, pipeline, background, scene_suffix): dir_name = f"{name}{scene_suffix}" render_path = os.path.join(model_path, dir_name, "ours_{}".format(iteration), "renders") hair_mask_path = os.path.join(model_path, dir_name, "ours_{}".format(iteration), "hair_masks") head_mask_path = os.path.join(model_path, dir_name, "ours_{}".format(iteration), "head_masks") orient_path = os.path.join(model_path, dir_name, "ours_{}".format(iteration), "orients") orient_vis_path = os.path.join(model_path, dir_name, "ours_{}".format(iteration), "orients_vis") orient_conf_path = os.path.join(model_path, dir_name, "ours_{}".format(iteration), "orient_confs") orient_conf_vis_path = os.path.join(model_path, dir_name, "ours_{}".format(iteration), "orient_confs_vis") makedirs(render_path, exist_ok=True) makedirs(hair_mask_path, exist_ok=True) makedirs(head_mask_path, exist_ok=True) makedirs(orient_path, exist_ok=True) makedirs(orient_vis_path, exist_ok=True) makedirs(orient_conf_path, exist_ok=True) makedirs(orient_conf_vis_path, exist_ok=True) for idx, view in enumerate(tqdm(views, desc="Rendering progress")): output = render(view, gaussians, pipeline, background) image = output["render"] hair_mask = output["mask"][:1] head_mask = output["mask"][1:] orient_angle = output["orient_angle"] * hair_mask orient_angle_vis = vis_orient(output["orient_angle"], hair_mask) orient_conf = output["orient_conf"] * hair_mask orient_conf_vis = (1 - 1 / (orient_conf + 1)) orient_conf_vis = vis_orient(output["orient_angle"], orient_conf_vis) basename = os.path.basename(view.image_name).split('.')[0] torchvision.utils.save_image(image, os.path.join(render_path, basename + ".png")) torchvision.utils.save_image(hair_mask, os.path.join(hair_mask_path, basename + ".png")) torchvision.utils.save_image(head_mask, os.path.join(head_mask_path, basename + ".png")) torchvision.utils.save_image(orient_angle, os.path.join(orient_path, basename + ".png")) torchvision.utils.save_image(orient_angle_vis, os.path.join(orient_vis_path, basename + ".png")) torch.save(orient_conf, os.path.join(orient_conf_path, basename + ".pth")) torchvision.utils.save_image(orient_conf_vis, os.path.join(orient_conf_vis_path, basename + ".png")) @torch.no_grad() def render_sets(dataset : ModelParams, optimizer: OptimizationParams, iteration : int, pipeline : PipelineParams, skip_train : bool, skip_test : bool, scene_suffix : str): gaussians = GaussianModel(dataset.sh_degree) scene = Scene(dataset, gaussians, load_iteration=iteration, shuffle=False, scene_suffix=scene_suffix) gaussians.training_setup(optimizer) if dataset.trainable_cameras: print(f'Loading optimized cameras from iter {scene.loaded_iter}') params_cam_rotation, params_cam_translation, params_cam_fov = pkl.load(open(scene.model_path + "/cameras/" + str(scene.loaded_iter) + ".pkl", 'rb')) for k in scene.train_cameras.keys(): for camera in scene.train_cameras[k]: if dataset.trainable_cameras: camera._rotation_res.data = params_cam_rotation[camera.image_name] camera._translation_res.data = params_cam_translation[camera.image_name] if dataset.trainable_intrinsics: camera._fov_res.data = params_cam_fov[camera.image_name] projection_all = {} params_all = {} for camera in scene.train_cameras[1.0]: projection_all[camera.image_name] = camera.full_proj_transform.cpu() params_all[camera.image_name] = { 'fx': fov2focal(camera.FoVx, camera.width).item(), 'fy': fov2focal(camera.FoVy, camera.height).item(), 'width': camera.width, 'height': camera.height, 'Rt': camera.world_view_transform.cpu().transpose(0, 1) } pkl.dump(projection_all, open(scene.model_path + "/cameras/" + str(scene.loaded_iter) + "_matrices.pkl", 'wb')) pkl.dump(params_all, open(scene.model_path + "/cameras/" + str(scene.loaded_iter) + "_params.pkl", 'wb')) 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") if not skip_train: render_set(dataset.model_path, "train", scene.loaded_iter, scene.getTrainCameras(), gaussians, pipeline, background, scene_suffix) if not skip_test: render_set(dataset.model_path, "test", scene.loaded_iter, scene.getTestCameras(), gaussians, pipeline, background, scene_suffix) if __name__ == "__main__": # Set up command line argument parser parser = ArgumentParser(description="Testing script parameters") model = ModelParams(parser, sentinel=True) optimizer = OptimizationParams(parser) pipeline = PipelineParams(parser) parser.add_argument("--iteration", default=-1, type=int) parser.add_argument("--data_dir", default="", type=str) parser.add_argument("--scene_suffix", default="", type=str) parser.add_argument("--skip_train", action="store_true") parser.add_argument("--skip_test", action="store_true") parser.add_argument("--quiet", action="store_true") args = get_combined_args(parser) print("Rendering " + args.model_path) # Initialize system state (RNG) safe_state(args.quiet) dataset = model.extract(args) if args.data_dir: dataset.source_path = args.data_dir render_sets(dataset, optimizer.extract(args), args.iteration, pipeline.extract(args), args.skip_train, args.skip_test, args.scene_suffix)