import os import sys import random import torch import numpy as np import torch.nn.functional as F from argparse import ArgumentParser from core.registry import register_method from core.base_method import BaseMethod sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), '../../gaussianpro_official'))) from utils.loss_utils import l1_loss, ssim from gaussian_renderer import render as native_render from scene import Scene, GaussianModel from arguments import ModelParams, PipelineParams, OptimizationParams @register_method("gaussianpro") class GaussianProWrapper(BaseMethod): def __init__(self, dataset_config, hyperparams): self.parser = ArgumentParser() self.lp = ModelParams(self.parser) self.op = OptimizationParams(self.parser) self.pp = PipelineParams(self.parser) self.args = self.parser.parse_args([]) self.args.source_path = dataset_config["source_path"] self.args.model_path = dataset_config["model_path"] self.args.eval = True self.args.resolution = dataset_config.get("resolution", 1) self.track_decoupling = hyperparams.get("track_decoupling", False) self.dataset = self.lp.extract(self.args) self.opt = self.op.extract(self.args) self.pipe = self.pp.extract(self.args) self.opt.normal_loss = True self.opt.flatten_loss = True self.opt.sparse_loss = True self.opt.lambda_flatten = 100.0 self.opt.lambda_l1_normal = 0.01 self.opt.lambda_cos_normal = 0.01 self.opt.lambda_sparse = 0.001 self.gaussians = GaussianModel(self.dataset.sh_degree) # INJECTED_RES_FIX begin import sys as _sys _scene, _explicit_res = None, None for _i, _a in enumerate(_sys.argv[:-1]): _v = _sys.argv[_i + 1] if _a == "--scene": _scene = _v elif _a == "--source_path": _scene = _v.rstrip("/").split("/")[-1] elif _a == "--resolution": try: _explicit_res = int(_v) except: pass _OUTDOOR_360 = {"bicycle", "flowers", "garden", "stump", "treehill"} if _explicit_res is not None and _explicit_res > 0: _res = _explicit_res elif _scene is not None: _res = 4 if _scene in _OUTDOOR_360 else 2 else: _res = None try: if _res is not None: self.dataset.resolution = _res print("[res-fix] scene=%s explicit=%s -> res=%s (%s)" % (_scene, _explicit_res, _res, __file__)) except Exception as _e: print("[res-fix] FAILED:", _e) # INJECTED_RES_FIX end self.scene = Scene(self.dataset, self.gaussians) self.gaussians.training_setup(self.opt) bg_color = [1, 1, 1] if self.dataset.white_background else [0, 0, 0] self.background = torch.tensor(bg_color, dtype=torch.float32, device="cuda") self.viewpoint_stack = self.scene.getTrainCameras().copy() self.last_n_gaussians = len(self.gaussians.get_xyz) def train_iteration(self, step): self.gaussians.update_learning_rate(step) if step % 1000 == 0: self.gaussians.oneupSHdegree() if not self.viewpoint_stack: self.viewpoint_stack = self.scene.getTrainCameras().copy() viewpoint_cam = self.viewpoint_stack.pop(random.randint(0, len(self.viewpoint_stack) - 1)) render_pkg = native_render(viewpoint_cam, self.gaussians, self.pipe, self.background, return_normal=self.opt.normal_loss, return_opacity=True) image = render_pkg["render"] viewspace_point_tensor = render_pkg["viewspace_points"] visibility_filter = render_pkg["visibility_filter"] radii = render_pkg["radii"] opacity_mask = render_pkg["render_opacity"] > 0.0 opacity_mask = opacity_mask.unsqueeze(0).expand(3, -1, -1) gt_image = viewpoint_cam.original_image.cuda() Ll1 = l1_loss(image[opacity_mask], gt_image[opacity_mask]) ssim_value = ssim(image, gt_image, mask=opacity_mask) loss_target = (1.0 - self.opt.lambda_dssim) * Ll1 loss_target_ssim = self.opt.lambda_dssim * (1.0 - ssim_value) loss_target_total = loss_target + loss_target_ssim loss_parasitic = torch.tensor(0.0, device="cuda") loss_flatten_val = 0.0 loss_normal_val = 0.0 if self.opt.flatten_loss: scales = self.gaussians.get_scaling min_scale, _ = torch.min(scales, dim=1) min_scale = torch.clamp(min_scale, 0, 30) flatten_loss_tensor = torch.abs(min_scale).mean() loss_parasitic = loss_parasitic + self.opt.lambda_flatten * flatten_loss_tensor loss_flatten_val = float(flatten_loss_tensor) if self.opt.sparse_loss: opacity = self.gaussians.get_opacity.clamp(1e-6, 1-1e-6) log_opacity = opacity * torch.log(opacity) log_one_minus_opacity = (1 - opacity) * torch.log(1 - opacity) sparse_loss_tensor = -1 * (log_opacity + log_one_minus_opacity)[visibility_filter].mean() loss_parasitic = loss_parasitic + self.opt.lambda_sparse * sparse_loss_tensor if self.opt.normal_loss and viewpoint_cam.normal is not None: rendered_normal = render_pkg["render_normal"] normal_gt = viewpoint_cam.normal.cuda() filter_mask = (normal_gt != -10)[0, :, :].to(torch.bool) if filter_mask.any(): l1_normal = torch.abs(rendered_normal - normal_gt).sum(dim=0)[filter_mask].mean() cos_normal = (1.0 - torch.sum(rendered_normal * normal_gt, dim=0))[filter_mask].mean() normal_loss_tensor = self.opt.lambda_l1_normal * l1_normal + self.opt.lambda_cos_normal * cos_normal loss_parasitic = loss_parasitic + normal_loss_tensor loss_normal_val = float(normal_loss_tensor) loss = loss_target_total + loss_parasitic grad_cos_sim = 0.0 parasitic_ratio = 0.0 if self.track_decoupling and step % 100 == 0 and loss_parasitic.requires_grad: track_params = [ self.gaussians._xyz, self.gaussians._scaling, self.gaussians._opacity, self.gaussians._rotation ] self.gaussians.optimizer.zero_grad(set_to_none=True) loss_target_total.backward(retain_graph=True) grad_t_list = [p.grad.clone().view(p.shape[0], -1) if p.grad is not None else torch.zeros_like(p).view(p.shape[0], -1) for p in track_params] grad_target_cat = torch.cat(grad_t_list, dim=1) self.gaussians.optimizer.zero_grad(set_to_none=True) loss_parasitic.backward(retain_graph=True) grad_p_list = [p.grad.clone().view(p.shape[0], -1) if p.grad is not None else torch.zeros_like(p).view(p.shape[0], -1) for p in track_params] grad_parasitic_cat = torch.cat(grad_p_list, dim=1) valid_mask = (torch.norm(grad_target_cat, dim=1) > 0) & (torch.norm(grad_parasitic_cat, dim=1) > 0) if valid_mask.any(): grad_cos_sim = float(F.cosine_similarity(grad_target_cat[valid_mask], grad_parasitic_cat[valid_mask], dim=1).mean()) parasitic_ratio = float(torch.norm(grad_parasitic_cat, dim=1).mean() / (torch.norm(grad_target_cat, dim=1).mean() + 1e-7)) self.gaussians.optimizer.zero_grad(set_to_none=True) loss.backward() else: loss.backward() with torch.no_grad(): if step < self.opt.densify_until_iter: self.gaussians.max_radii2D[visibility_filter] = torch.max(self.gaussians.max_radii2D[visibility_filter], radii[visibility_filter]) self.gaussians.add_densification_stats(viewspace_point_tensor, visibility_filter) if step > self.opt.densify_from_iter and step % self.opt.densification_interval == 0: size_threshold = 20 if step > self.opt.opacity_reset_interval else None self.gaussians.densify_and_prune(self.opt.densify_grad_threshold, 0.005, self.scene.cameras_extent, size_threshold) if step % self.opt.opacity_reset_interval == 0 or (self.dataset.white_background and step == self.opt.densify_from_iter): self.gaussians.reset_opacity() self.gaussians.optimizer.step() self.gaussians.optimizer.zero_grad(set_to_none=True) num_gaussians = self.gaussians.get_xyz.shape[0] metrics = { "loss": float(loss), "loss_l1": float(loss_target), "loss_ssim": float(loss_target_ssim), "loss_flatten": loss_flatten_val, "loss_normal": loss_normal_val, "num_gaussians": int(num_gaussians), "delta_N": int(num_gaussians - self.last_n_gaussians), "peak_vram_GB": float(torch.cuda.max_memory_allocated() / (1024 ** 3)), "grad_cos_sim": float(grad_cos_sim), "parasitic_ratio": float(parasitic_ratio) } self.last_n_gaussians = num_gaussians histograms = {} if step % 1000 == 0: histograms["opacity"] = torch.sigmoid(self.gaussians._opacity).clone().detach() scales = torch.exp(self.gaussians._scaling).clone().detach() histograms["scaling"] = scales max_S, _ = torch.max(scales, dim=-1) min_S, _ = torch.min(scales, dim=-1) histograms["anisotropy"] = max_S / (min_S + 1e-7) histograms["sh_dc_mag"] = self.gaussians._features_dc.detach().norm(dim=-1) return metrics, histograms def render(self, camera): with torch.no_grad(): render_pkg = native_render(camera, self.gaussians, self.pipe, self.background) return {"image": render_pkg["render"], "depth": render_pkg.get("render_depth", None)} def save(self, save_dir, step): self.scene.save(step) def load(self, model_path, iteration): self.gaussians.load_ply(os.path.join(model_path, "point_cloud", f"iteration_{iteration}", "point_cloud.ply")) def get_spatial_centers(self): return self.gaussians._xyz def compute_physical_metrics(self, cameras=None): metrics = {} with torch.no_grad(): scales = torch.exp(self.gaussians._scaling) max_S, _ = torch.max(scales, dim=1) min_S, _ = torch.min(scales, dim=1) gamma = max_S / (min_S + 1e-7) metrics["gamma_median"] = float(torch.median(gamma)) metrics["gamma_90th_percentile"] = float(torch.quantile(gamma, 0.90)) metrics["scale_mean"] = float(torch.mean(scales)) metrics["alpha_mean"] = float(torch.mean(torch.sigmoid(self.gaussians._opacity))) dc, rest = self.gaussians._features_dc, self.gaussians._features_rest if rest is not None and rest.shape[1] > 0: metrics["sh_energy_ratio"] = float(rest.norm(dim=-1).mean() / (dc.norm(dim=-1).mean() + 1e-7)) if cameras is not None and len(cameras) > 0: view_dirs = [] for c in cameras: view_dirs.append(c.world_view_transform[:3, 2].tolist()) view_dirs = F.normalize(torch.tensor(view_dirs, dtype=torch.float32, device="cuda"), dim=1) rots = F.normalize(self.gaussians._rotation.clone(), dim=1) w, x, y, z = rots.unbind(dim=-1) normals = F.normalize(torch.stack([2*(x*z + w*y), 2*(y*z - w*x), 1-2*(x*x + y*y)], dim=-1), dim=1) max_cos, _ = torch.max(torch.abs(torch.matmul(normals, view_dirs.T)), dim=1) metrics["billboard_bias_ratio"] = float((max_cos > 0.90).float().mean()) return metrics def evaluate_spatial_field(self, query_points: torch.Tensor, cameras=None) -> torch.Tensor: with torch.no_grad(): V = query_points.shape[0] densities = torch.zeros(V, device="cuda") xyz, opacities = self.gaussians._xyz, torch.sigmoid(self.gaussians._opacity).squeeze() scales = torch.exp(self.gaussians._scaling) sigma_sq = scales.max(dim=1)[0].pow(2) N_gaussians = xyz.shape[0] chunk_size = max(1, 30_000_000 // (N_gaussians + 1)) for i in range(0, V, chunk_size): end = min(i + chunk_size, V) dist_sq = torch.cdist(query_points[i:end], xyz, p=2).pow(2) weights = torch.exp(-0.5 * dist_sq / (sigma_sq.unsqueeze(0) + 1e-7)) densities[i:end] = torch.sum(weights * opacities.unsqueeze(0), dim=1) return densities