| 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__), '../../erank_gs'))) |
| 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 |
| from utils.get_stats import get_effective_rank, get_ordered_scale_multiple |
|
|
| @register_method("erankgs") |
| class ERankGSWrapper(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.erank_from_iter = 7000 |
| self.opt.no_new_densification = False |
|
|
| self.gaussians = GaussianModel(self.dataset.sh_degree) |
| |
| 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) |
| |
| 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)) |
| |
| bg = torch.rand((3), device="cuda") if self.opt.random_background else self.background |
| render_pkg = native_render(viewpoint_cam, self.gaussians, self.pipe, bg, return_normal=True, return_depth=True) |
| image = render_pkg["render"] |
| viewspace_point_tensor = render_pkg["viewspace_points"] |
| visibility_filter = render_pkg["visibility_filter"] |
| radii = render_pkg["radii"] |
| |
| gt_image = viewpoint_cam.original_image.cuda() |
| |
| Ll1 = l1_loss(image, gt_image) |
| ssim_value = ssim(image, gt_image) |
| loss_target = (1.0 - self.opt.lambda_dssim) * Ll1 + self.opt.lambda_dssim * (1.0 - ssim_value) |
| |
| erank_loss = torch.tensor(0.0, device="cuda") |
| thin_loss = torch.tensor(0.0, device="cuda") |
| |
| scales = self.gaussians.get_scaling |
| erank = get_effective_rank(scales) |
| |
| if self.opt.erank_from_iter <= step <= self.opt.erank_end_iter: |
| erank_loss = self.opt.erank_lambda * torch.clamp(-torch.log(erank - 1 + 1e-7), 0).mean() |
| _, ordered_scale = get_ordered_scale_multiple(scales) |
| thin_loss = self.opt.thin_lambda * ordered_scale[:, 2].mean() |
| |
| loss_parasitic = erank_loss + thin_loss |
| loss = loss_target + loss_parasitic |
|
|
| image_mask = viewpoint_cam.image_mask |
| if image_mask is not None: |
| o = render_pkg['render_opacity'] |
| opacity_mask_loss = -(image_mask * torch.log(o + 1e-7) + (1 - image_mask) * torch.log(1 - o + 1e-7)).mean() |
| loss += opacity_mask_loss |
|
|
| grad_cos_sim = 0.0 |
| parasitic_ratio = 0.0 |
|
|
| if self.track_decoupling and step % 100 == 0 and loss_parasitic.item() > 0: |
| self.gaussians.optimizer.zero_grad(set_to_none=True) |
| loss_target.backward(retain_graph=True) |
| grad_target = self.gaussians._scaling.grad.clone() if self.gaussians._scaling.grad is not None else torch.zeros_like(self.gaussians._scaling) |
| |
| self.gaussians.optimizer.zero_grad(set_to_none=True) |
| loss_parasitic.backward(retain_graph=True) |
| grad_parasitic = self.gaussians._scaling.grad.clone() if self.gaussians._scaling.grad is not None else torch.zeros_like(self.gaussians._scaling) |
| |
| lr_scaling = 0.005 |
| v_t = torch.zeros_like(self.gaussians._scaling) |
| for group in self.gaussians.optimizer.param_groups: |
| if group["name"] == "scaling": |
| lr_scaling = group["lr"] |
| state = self.gaussians.optimizer.state.get(group['params'][0], None) |
| if state is not None and "exp_avg_sq" in state: |
| v_t = state["exp_avg_sq"].clone() |
| break |
|
|
| u_t = (lr_scaling / (torch.sqrt(v_t) + 1e-15)) * grad_target |
| u_p = (lr_scaling / (torch.sqrt(v_t) + 1e-15)) * grad_parasitic |
| |
| valid_mask = (torch.norm(u_t, dim=1) > 0) & (torch.norm(u_p, dim=1) > 0) |
| if valid_mask.any(): |
| grad_cos_sim = float(F.cosine_similarity(u_t[valid_mask], u_p[valid_mask], dim=1).mean()) |
| parasitic_ratio = float(torch.norm(u_p, dim=1).mean() / (torch.norm(u_t, 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 |
| if self.opt.no_new_densification: |
| self.gaussians.densify_and_prune(self.opt.densify_grad_threshold, 0.005, self.scene.cameras_extent, size_threshold) |
| else: |
| self.gaussians.densify_and_prune_v2(self.opt.densify_grad_threshold, 0.05, 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(Ll1), |
| "loss_ssim": float(ssim_value), |
| "loss_erank": float(erank_loss), |
| "erank_mean": float(erank.mean()), |
| "needle_ratio": float((erank < 1.04).float().mean()), |
| "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() |
| current_scales = torch.exp(self.gaussians._scaling).clone().detach() |
| histograms["scaling"] = current_scales |
| scales_2d = current_scales[:, :2] if current_scales.shape[1] >= 2 else current_scales |
| gamma = scales_2d.max(dim=-1)[0] / (scales_2d.min(dim=-1)[0] + 1e-7) |
| histograms["anisotropy"] = gamma |
| 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_depth=True) |
| 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(): |
| raw_scales = self.gaussians._scaling |
| scales = torch.exp(raw_scales) |
| |
| scales_2d = scales[:, :2] if scales.dim() > 1 and scales.shape[1] >= 2 else scales.unsqueeze(-1).expand(-1, 2) |
| |
| max_S, _ = torch.max(scales_2d, dim=1) |
| min_S, _ = torch.min(scales_2d, 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_2d)) |
| 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[:, :2].max(dim=1)[0].pow(2)) if scales.shape[1] >= 2 else scales.squeeze().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 |
|
|
