methods/wrapper_gsdf.py

import os
import sys
import random
import torch
import numpy as np
import torch.nn.functional as F
from argparse import ArgumentParser

# 强制路径隔离：将 GSDF 路径插入最前端并清理冲突模块
base_path = os.path.abspath(os.path.join(os.path.dirname(__file__), '../../GSDF_run'))
gs_path = os.path.join(base_path, "gaussian_splatting")
sys.path.insert(0, base_path)
sys.path.insert(0, gs_path)

for mod in ["gaussian_renderer", "scene", "arguments", "utils", "gaussian_renderer.render"]:
    if mod in sys.modules:
        del sys.modules[mod]

from core.registry import register_method
from core.base_method import BaseMethod

from utils.loss_utils import l1_loss, ssim
from gaussian_renderer import render as native_render
from gaussian_renderer import prefilter_voxel, generate_neural_gaussians
from scene import Scene, GaussianModel
from arguments import ModelParams, PipelineParams, OptimizationParams

@register_method("gsdf")
class GSDFWrapper(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.gaussians = GaussianModel(self.dataset.feat_dim, self.dataset.n_offsets, self.dataset.voxel_size, self.dataset.update_depth, self.dataset.update_init_factor, self.dataset.update_hierachy_factor, self.dataset.use_feat_bank, self.dataset.use_tcnn)
        # 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_anchors = self.gaussians.get_anchor.shape[0]
        
        self.cached_xyz = None
        self.cached_scaling = None
        self.cached_opacity = None
        self.cached_rot = None

    def train_iteration(self, step):
        self.gaussians.update_learning_rate(step)
            
        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))
        
        voxel_visible_mask = prefilter_voxel(viewpoint_cam, self.gaussians, self.pipe, self.background)
        retain_grad = (step < 15000 and step >= 0)
        
        render_pkg = native_render(viewpoint_cam, self.gaussians, self.pipe, self.background, visible_mask=voxel_visible_mask, retain_grad=retain_grad)
        
        image = render_pkg["render"]
        viewspace_point_tensor = render_pkg["viewspace_points"]
        if retain_grad and viewspace_point_tensor is not None:
            viewspace_point_tensor.retain_grad()
        visibility_filter = render_pkg["visibility_filter"]
        offset_selection_mask = render_pkg["selection_mask"]
        scaling = render_pkg["scaling"]
        opacity = render_pkg["neural_opacity"]
        
        self.cached_xyz = render_pkg.get("xyz", None)
        self.cached_scaling = scaling
        self.cached_opacity = opacity
        self.cached_rot = render_pkg.get("rotations", None)
        
        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)
        
        loss_parasitic = 0.01 * scaling.prod(dim=1).mean()
        loss = loss_target + loss_parasitic

        grad_cos_sim = 0.0
        parasitic_ratio = 0.0

        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._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)
            
            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))
            
            self.gaussians.optimizer.zero_grad(set_to_none=True)
            loss.backward()
        else:
            loss.backward()

        with torch.no_grad():
            if step < self.opt.update_until and step > self.opt.start_stat:
                self.gaussians.training_statis(viewspace_point_tensor, opacity, visibility_filter, offset_selection_mask, voxel_visible_mask)
                if step > self.opt.update_from and step % self.opt.update_interval == 0:
                    self.gaussians.adjust_anchor(extent=self.scene.cameras_extent, check_interval=self.opt.update_interval, success_threshold=self.opt.success_threshold, grad_threshold=self.opt.densify_grad_threshold, min_opacity=self.opt.min_opacity)
            elif step == self.opt.update_until:
                if hasattr(self.gaussians, 'opacity_accum'): del self.gaussians.opacity_accum
                if hasattr(self.gaussians, 'offset_gradient_accum'): del self.gaussians.offset_gradient_accum
                if hasattr(self.gaussians, 'offset_denom'): del self.gaussians.offset_denom
                torch.cuda.empty_cache()

            if step < self.opt.iterations:
                self.gaussians.optimizer.step()
            self.gaussians.optimizer.zero_grad(set_to_none=True)

        num_anchors = self.gaussians.get_anchor.shape[0]
        num_gaussians = scaling.shape[0] if scaling is not None else 0
        
        metrics = {
            "loss": float(loss), 
            "loss_l1": float(loss_target), 
            "loss_ssim": float(loss_parasitic),
            "num_gaussians": int(num_gaussians), 
            "num_anchors": int(num_anchors),
            "delta_N": int(num_anchors - self.last_n_anchors),
            "peak_vram_GB": float(torch.cuda.max_memory_allocated() / (1024 ** 3)),
            "grad_cos_sim": float(grad_cos_sim), 
            "parasitic_ratio": float(parasitic_ratio),
            "anchor_occupancy_ratio": float(num_gaussians / (num_anchors * self.dataset.n_offsets + 1e-7)),
            "loss_scaling_reg": float(loss_parasitic)
        }
        self.last_n_anchors = num_anchors
        
        histograms = {}
        if step % 1000 == 0 and self.cached_scaling is not None:
            histograms["opacity"] = self.cached_opacity.clone().detach()
            histograms["scaling"] = self.cached_scaling.clone().detach()
            scales_2d = self.cached_scaling[:, :2] if self.cached_scaling.shape[1] >= 2 else self.cached_scaling
            gamma = scales_2d.max(dim=-1)[0] / (scales_2d.min(dim=-1)[0] + 1e-7)
            histograms["anisotropy"] = gamma

        return metrics, histograms
        
    def render(self, camera):
        with torch.no_grad():
            voxel_visible_mask = prefilter_voxel(camera, self.gaussians, self.pipe, self.background)
            render_pkg = native_render(camera, self.gaussians, self.pipe, self.background, visible_mask=voxel_visible_mask)
        return {"image": render_pkg["render"], "depth": render_pkg.get("depth_hand", None)}

    def save(self, save_dir, step):
        self.scene.save(step)

    def load(self, model_path, iteration):
        self.gaussians.load_ply_sparse_gaussian(os.path.join(model_path, 'point_cloud', f'iteration_{iteration}', 'point_cloud.ply'))
        self.gaussians.load_mlp_checkpoints(os.path.join(model_path, 'point_cloud', f'iteration_{iteration}', 'checkpoint.pth'))

    def get_spatial_centers(self):
        return self.gaussians.get_anchor

    def compute_physical_metrics(self, cameras=None):
        metrics = {}
        if self.cached_scaling is None or self.cached_opacity is None:
            return metrics
            
        with torch.no_grad():
            scales = self.cached_scaling
            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(self.cached_opacity))
            
            if cameras is not None and len(cameras) > 0 and self.cached_rot is not None:
                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.cached_rot.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")
            
            if cameras is None or len(cameras) == 0:
                cam = self.scene.getTrainCameras()[0]
            else:
                cam = cameras[0]
                
            xyz, _, opacity, scaling, _, _, _ = generate_neural_gaussians(cam, self.gaussians, None, is_training=True)
            opacity = opacity.squeeze()
            
            sigma_sq = (scaling[:, :2].max(dim=1)[0].pow(2)) if scaling.shape[1] >= 2 else scaling.squeeze().pow(2)
            
            N_gaussians = xyz.shape[0]
            if N_gaussians == 0:
                return densities
                
            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 * opacity.unsqueeze(0), dim=1)
            return densities