| import os |
| import sys |
| import random |
| import math |
| import torch |
| import numpy as np |
| import kornia |
| import nvdiffrast.torch as dr |
| 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__), '../../GI-GS_offy'))) |
|
|
| from utils.loss_utils import l1_loss, ssim |
| from utils.image_utils import psnr |
| from gaussian_renderer import render |
| from diff_gaussian_rasterization import Gaussian_SSR |
| from pbr import CubemapLight, get_brdf_lut, pbr_shading |
| from scene import Scene, GaussianModel |
| from arguments import ModelParams, PipelineParams, OptimizationParams |
|
|
| def linear_to_srgb(linear): |
| eps = torch.finfo(torch.float32).eps |
| srgb0 = 323 / 25 * linear |
| srgb1 = (211 * torch.clamp(linear, min=eps) ** (5 / 12) - 11) / 200 |
| return torch.where(linear <= 0.0031308, srgb0, srgb1) |
|
|
| def srgb_to_linear(srgb): |
| linear0 = 25 / 323 * srgb |
| linear1 = ((srgb + 0.055) / 1.055)**2.4 |
| return torch.where(srgb <= 0.04045, linear0, linear1) |
|
|
| def get_tv_loss(gt_image, prediction, pad=1, step=1): |
| if pad > 1: |
| gt_image = F.avg_pool2d(gt_image, pad, pad) |
| prediction = F.avg_pool2d(prediction, pad, pad) |
| rgb_grad_h = torch.exp(-(gt_image[:, 1:, :] - gt_image[:, :-1, :]).abs().mean(dim=0, keepdim=True)) |
| rgb_grad_w = torch.exp(-(gt_image[:, :, 1:] - gt_image[:, :, :-1]).abs().mean(dim=0, keepdim=True)) |
| tv_h = torch.pow(prediction[:, 1:, :] - prediction[:, :-1, :], 2) |
| tv_w = torch.pow(prediction[:, :, 1:] - prediction[:, :, :-1], 2) |
| tv_loss = (tv_h * rgb_grad_h).mean() + (tv_w * rgb_grad_w).mean() |
| if step > 1: |
| for s in range(2, step + 1): |
| rgb_grad_h = torch.exp(-(gt_image[:, s:, :] - gt_image[:, :-s, :]).abs().mean(dim=0, keepdim=True)) |
| rgb_grad_w = torch.exp(-(gt_image[:, :, s:] - gt_image[:, :, :-s]).abs().mean(dim=0, keepdim=True)) |
| tv_h = torch.pow(prediction[:, s:, :] - prediction[:, :-s, :], 2) |
| tv_w = torch.pow(prediction[:, :, s:] - prediction[:, :, :-s], 2) |
| tv_loss += (tv_h * rgb_grad_h).mean() + (tv_w * rgb_grad_w).mean() |
| return tv_loss |
|
|
| def get_masked_tv_loss(mask, gt_image, prediction): |
| rgb_grad_h = torch.exp(-(gt_image[:, 1:, :] - gt_image[:, :-1, :]).abs().mean(dim=0, keepdim=True)) |
| rgb_grad_w = torch.exp(-(gt_image[:, :, 1:] - gt_image[:, :, :-1]).abs().mean(dim=0, keepdim=True)) |
| tv_h = torch.pow(prediction[:, 1:, :] - prediction[:, :-1, :], 2) |
| tv_w = torch.pow(prediction[:, :, 1:] - prediction[:, :, :-1], 2) |
| mask = mask.float() |
| mask_h = mask[:, 1:, :] * mask[:, :-1, :] |
| mask_w = mask[:, :, 1:] * mask[:, :, :-1] |
| tv_loss = (tv_h * rgb_grad_h * mask_h).mean() + (tv_w * rgb_grad_w * mask_w).mean() |
| return tv_loss |
|
|
| def get_envmap_dirs(res=[512, 1024]): |
| gy, gx = torch.meshgrid( |
| torch.linspace(0.0 + 1.0 / res[0], 1.0 - 1.0 / res[0], res[0], device="cuda"), |
| torch.linspace(-1.0 + 1.0 / res[1], 1.0 - 1.0 / res[1], res[1], device="cuda"), |
| indexing="ij", |
| ) |
| sintheta, costheta = torch.sin(gy * np.pi), torch.cos(gy * np.pi) |
| sinphi, cosphi = torch.sin(gx * np.pi), torch.cos(gx * np.pi) |
| reflvec = torch.stack((sintheta * sinphi, costheta, -sintheta * cosphi), dim=-1) |
| return reflvec |
|
|
| @register_method("gigs") |
| class GIGSWrapper(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.resolution = dataset_config.get("resolution", 1) |
| self.args.eval = True |
| self.args.metallic = True |
| self.args.indirect = True |
| self.args.pbr_iteration = 7000 |
| self.args.tone = False |
| self.args.gamma = False |
| self.args.normal_tv = 1.0 |
| self.args.brdf_tv = 1.0 |
| self.args.env_tv = 0.01 |
| self.args.radius = 0.8 |
| self.args.bias = 0.01 |
| self.args.thick = 0.05 |
| self.args.delta = 0.0625 |
| self.args.step = 16 |
| self.args.start = 8 |
|
|
| if "360" in str(self.args.source_path).lower() and "indoor" not in str(self.args.source_path).lower(): |
| self.args.degree = 3 |
| else: |
| self.args.degree = 1 |
|
|
| self.track_decoupling = hyperparams.get("track_decoupling", False) |
|
|
| self.dataset = self.lp.extract(self.args) |
| self.dataset.sh_degree = self.args.degree |
| self.opt = self.op.extract(self.args) |
| self.pipe = self.pp.extract(self.args) |
|
|
| 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) |
|
|
| self.brdf_lut = get_brdf_lut().cuda() |
| self.envmap_dirs = get_envmap_dirs() |
| self.cubemap = CubemapLight(base_res=256).cuda() |
| self.cubemap.train() |
|
|
| param_groups = [{"name": "cubemap", "params": self.cubemap.parameters(), "lr": self.opt.opacity_lr}] |
| self.light_optimizer = torch.optim.Adam(param_groups, lr=self.opt.opacity_lr) |
|
|
| 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) |
| self.canonical_rays = self.scene.get_canonical_rays() |
|
|
| 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)) |
|
|
| try: |
| c2w = torch.inverse(viewpoint_cam.world_view_transform.T) |
| except: |
| return {}, {} |
|
|
| bg = torch.rand((3), device="cuda") if self.opt.random_background else self.background |
| current_bg = bg if step <= self.args.pbr_iteration else torch.zeros_like(bg) |
|
|
| render_pkg = render( |
| viewpoint_camera=viewpoint_cam, |
| pc=self.gaussians, |
| pipe=self.pipe, |
| bg_color=current_bg, |
| pad_normal=False, |
| derive_normal=True, |
| radius=self.args.radius, |
| bias=self.args.bias, |
| thick=self.args.thick, |
| delta=self.args.delta, |
| step=self.args.step, |
| start=self.args.start |
| ) |
|
|
| image = render_pkg["render"] |
| viewspace_point_tensor = render_pkg["viewspace_points"] |
| visibility_filter = render_pkg["visibility_filter"] |
| radii = render_pkg["radii"] |
| normal_map_from_depth = render_pkg["normal_map_from_depth"] |
| normal_map = render_pkg["normal_map"] |
| albedo_map = render_pkg["albedo_map"] |
| roughness_map = render_pkg["roughness_map"] |
| metallic_map = render_pkg["metallic_map"] |
|
|
| rmax, rmin = 1.0, 0.04 |
| roughness_map = roughness_map * (rmax - rmin) + rmin |
|
|
| H, W = viewpoint_cam.image_height, viewpoint_cam.image_width |
| view_dirs = -((F.normalize(self.canonical_rays[:, None, :], p=2, dim=-1) * c2w[None, :3, :3]).sum(dim=-1).reshape(H, W, 3)) |
|
|
| alpha_mask = viewpoint_cam.gt_alpha_mask.cuda() |
| gt_image = viewpoint_cam.original_image[0:3, :, :].cuda() |
| gt_image = (gt_image * alpha_mask + current_bg[:, None, None] * (1.0 - alpha_mask)).clamp(0.0, 1.0) |
|
|
| loss_target = torch.tensor(0.0, device="cuda") |
| loss_parasitic = torch.tensor(0.0, device="cuda") |
|
|
| if step <= self.args.pbr_iteration: |
| Ll1 = F.l1_loss(image, gt_image) |
| loss_target = (1.0 - self.opt.lambda_dssim) * Ll1 |
| ssim_val = ssim(image, gt_image) |
| mask = render_pkg["normal_from_depth_mask"] |
| normal_loss = F.l1_loss(normal_map[:, mask], normal_map_from_depth[:, mask]) |
| normal_tv_loss = get_tv_loss(gt_image, normal_map, pad=1, step=1) |
| loss_parasitic = self.opt.lambda_dssim * (1.0 - ssim_val) + normal_loss + normal_tv_loss * self.args.normal_tv |
| loss = loss_target + loss_parasitic |
| else: |
| occlusion = render_pkg["occlusion_map"].permute(1, 2, 0) if self.args.indirect else torch.ones_like(roughness_map).permute(1, 2, 0) |
| normal_mask = render_pkg["normal_mask"] |
| out_normal_view = render_pkg["out_normal_view"] |
| depth_pos = render_pkg["depth_pos"] |
|
|
| self.cubemap.build_mips() |
| pbr_result = pbr_shading( |
| light=self.cubemap, |
| normals=normal_map.permute(1, 2, 0).detach(), |
| view_dirs=view_dirs, |
| mask=normal_mask.permute(1, 2, 0), |
| albedo=albedo_map.permute(1, 2, 0), |
| roughness=roughness_map.permute(1, 2, 0), |
| metallic=metallic_map.permute(1, 2, 0) if self.args.metallic else None, |
| tone=self.args.tone, |
| gamma=self.args.gamma, |
| occlusion=occlusion.detach(), |
| brdf_lut=self.brdf_lut |
| ) |
|
|
| diffuse_rgb = pbr_result["diffuse_rgb"].clamp(min=0.0, max=1.0).permute(2, 0, 1) |
| diffuse_rgb = torch.where(normal_mask, diffuse_rgb, current_bg[:, None, None]) |
| render_direct = pbr_result["render_rgb"].permute(2, 0, 1) |
| render_direct = torch.where(normal_mask, render_direct, current_bg[:, None, None]) |
|
|
| tanfovx = math.tan(viewpoint_cam.FoVx * 0.5) |
| tanfovy = math.tan(viewpoint_cam.FoVy * 0.5) |
| SSR = Gaussian_SSR(tanfovx, tanfovy, W, H, self.args.radius, self.args.bias, self.args.thick, self.args.delta, self.args.step, self.args.start) |
| |
| if self.args.metallic: |
| F0 = (1.0 - metallic_map) * 0.04 + albedo_map * metallic_map |
| else: |
| F0 = torch.ones_like(albedo_map) * 0.04 |
| metallic_map = torch.zeros_like(roughness_map) |
| |
| linear_rgb = srgb_to_linear(render_direct) |
| (IRR, _) = SSR(out_normal_view.detach(), depth_pos.detach(), linear_rgb.detach(), albedo_map, roughness_map, metallic_map, F0) |
| IRR = linear_to_srgb(IRR) |
| IRR = kornia.filters.median_blur(IRR[None, ...], (3, 3))[0] |
| render_rgb = render_direct + IRR |
|
|
| loss_target = l1_loss(render_rgb, gt_image) |
|
|
| if (normal_mask == 0).sum() > 0: |
| brdf_tv_loss = get_masked_tv_loss(normal_mask, gt_image, torch.cat([albedo_map, roughness_map, metallic_map], dim=0)) |
| else: |
| brdf_tv_loss = get_tv_loss(gt_image, torch.cat([albedo_map, roughness_map, metallic_map], dim=0), pad=1, step=1) |
|
|
| lamb_loss = (1.0 - roughness_map[normal_mask]).mean() + metallic_map[normal_mask].mean() |
| envmap = dr.texture(self.cubemap.base[None, ...], self.envmap_dirs[None, ...].contiguous(), filter_mode="linear", boundary_mode="cube")[0] |
| tv_h1 = torch.pow(envmap[1:, :, :] - envmap[:-1, :, :], 2).mean() |
| tv_w1 = torch.pow(envmap[:, 1:, :] - envmap[:, :-1, :], 2).mean() |
| env_tv_loss = tv_h1 + tv_w1 |
|
|
| loss_parasitic = brdf_tv_loss * self.args.brdf_tv + lamb_loss * 0.001 + env_tv_loss * self.args.env_tv |
| loss = loss_target + loss_parasitic |
|
|
| grad_cos_sim = 0.0 |
| parasitic_ratio = 0.0 |
| stats = {} |
|
|
| if self.track_decoupling and step % 100 == 0: |
| self.gaussians.optimizer.zero_grad(set_to_none=True) |
| loss_target.backward(retain_graph=True) |
| grad_target = self.gaussians._xyz.grad.clone() if self.gaussians._xyz.grad is not None else torch.zeros_like(self.gaussians._xyz) |
| |
| self.gaussians.optimizer.zero_grad(set_to_none=True) |
| loss_parasitic.backward(retain_graph=True) |
| grad_parasitic = self.gaussians._xyz.grad.clone() if self.gaussians._xyz.grad is not None else torch.zeros_like(self.gaussians._xyz) |
| |
| valid_mask = (torch.norm(grad_target, dim=1) > 0) & (torch.norm(grad_parasitic, dim=1) > 0) |
| if valid_mask.any(): |
| grad_cos_sim = float(F.cosine_similarity(grad_target[valid_mask], grad_parasitic[valid_mask], dim=1).mean()) |
| parasitic_ratio = float(torch.norm(grad_parasitic, dim=1).mean() / (torch.norm(grad_target, dim=1).mean() + 1e-7)) |
| |
| param_groups_map = { |
| "spatial": [self.gaussians._xyz], |
| "geometry": [self.gaussians._scaling, self.gaussians._rotation], |
| "opacity": [self.gaussians._opacity], |
| "appearance": [self.gaussians._features_dc, self.gaussians._features_rest, self.cubemap.base], |
| } |
|
|
| self.gaussians.optimizer.zero_grad(set_to_none=True) |
| if step > self.args.pbr_iteration: |
| self.light_optimizer.zero_grad(set_to_none=True) |
|
|
| loss_target.backward(retain_graph=True) |
| grads_target = {} |
| for group_name, params in param_groups_map.items(): |
| grads_target[group_name] = torch.cat([p.grad.clone().reshape(-1) for p in params if p is not None and p.grad is not None]) |
|
|
| self.gaussians.optimizer.zero_grad(set_to_none=True) |
| if step > self.args.pbr_iteration: |
| self.light_optimizer.zero_grad(set_to_none=True) |
|
|
| loss_parasitic.backward(retain_graph=True) |
| grads_parasitic = {} |
| for group_name, params in param_groups_map.items(): |
| grads_parasitic[group_name] = torch.cat([p.grad.clone().reshape(-1) for p in params if p is not None and p.grad is not None]) |
|
|
| for group_name in param_groups_map: |
| gt, gp = grads_target.get(group_name), grads_parasitic.get(group_name) |
| if gt is not None and gp is not None and gt.numel() > 0 and gp.numel() > 0 and gt.norm() > 0 and gp.norm() > 0: |
| cos = float(F.cosine_similarity(gt.unsqueeze(0), gp.unsqueeze(0))) |
| r = float(gp.norm() / (gt.norm() + gp.norm() + 1e-7)) |
| ti = r * max(0.0, -cos) |
| else: |
| ti = 0.0 |
| stats[f"sti_{group_name}"] = ti |
|
|
| self.gaussians.optimizer.zero_grad(set_to_none=True) |
| if step > self.args.pbr_iteration: |
| self.light_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, 0.05, 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) |
|
|
| if step >= self.args.pbr_iteration: |
| self.light_optimizer.step() |
| self.light_optimizer.zero_grad(set_to_none=True) |
| self.cubemap.clamp_(min=0.0) |
|
|
| num_gaussians = self.gaussians.get_xyz.shape[0] |
| metrics = { |
| "loss": float(loss), "loss_l1": float(loss_target), "loss_ssim": float(loss_parasitic), |
| "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), |
| "cubemap_energy": float(self.cubemap.base.norm(p=2)), |
| "roughness_mean": float(roughness_map.mean()), |
| "metallic_ratio": float((metallic_map > 0.5).float().mean()) |
| } |
| metrics.update(stats) |
| 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 |
| scales_2d = scales[:, :2] if scales.shape[1] >= 2 else 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 = render(camera, self.gaussians, self.pipe, self.background, pad_normal=False, derive_normal=True, radius=self.args.radius, bias=self.args.bias, thick=self.args.thick, delta=self.args.delta, step=self.args.step, start=self.args.start) |
| return {"image": render_pkg["render"], "depth": render_pkg.get("depth_map", None)} |
|
|
| def save(self, save_dir, step): |
| self.scene.save(step) |
| torch.save({"cubemap": self.cubemap.state_dict(), "light_optimizer": self.light_optimizer.state_dict(), "iteration": step}, os.path.join(save_dir, f"chkpnt{step}.pth")) |
|
|
| def load(self, model_path, iteration): |
| self.gaussians.load_ply(os.path.join(model_path, 'point_cloud', f'iteration_{iteration}', 'point_cloud.ply')) |
| chkpnt = os.path.join(model_path, f"chkpnt{iteration}.pth") |
| if os.path.exists(chkpnt): |
| ckpt = torch.load(chkpnt) |
| self.cubemap.load_state_dict(ckpt["cubemap"]) |
|
|
| 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 |
|
|
|
|
