Research map / 3D vision / 3D generation
HAD: Hallucination-Aware Diffusion Priors for 3D Reconstruction
Xi Liu, Weiwei Sun, Zhou Ren, Chris Broaddus, Siyu Huang, Laurent Guigues
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Difix3D diffusion prior with hallucination-aware prior (HAD) (ablation) | 3DGS pipeline · DL3DV (9 sparse views) | PSNR: 21.355 -> 21.779 | Improved |
| add multi-sampling (3 versions) to HAD (ablation) | HAD · DL3DV (9 sparse views) | PSNR: 21.779 -> 21.983 | Improved |
| use single-phase 3DGS training with HAD, full model (ablation) | HAD · DL3DV (9 sparse views) | PSNR: 21.983 -> 22.134 | Improved |
| increase multi-sampling versions from 1 to 2 (ablation) | HAD · DL3DV (9 sparse views) | PSNR: 21.779 -> 21.938 | Improved |
| increase multi-sampling versions from 2 to 3 (ablation) | HAD · DL3DV (9 sparse views) | PSNR: 21.938 -> 21.983 | Improved |
| ArgMin fusion instead of weighted average (ablation) | HAD multi-sampling · DL3DV (9 sparse views) | PSNR: 21.856 -> 22.134 | Improved |
| use pretrained multiview encoder in score network (ablation) | hallucination score network · Score estimation test scenes | MAE: 0.054 -> 0.043 | Improved |
| use LVSM-based score network rather than retrained Difix3D estimator (ablation) | hallucination score network · Score estimation test scenes | MAE: 0.058 -> 0.043 | Improved |
| apply HAD in dense-view setting | Difix3D (Gsplat-MCMC) · DL3DV (24 dense views) | PSNR: 26.770 -> 26.969 | Improved |
| replace Difix3D with HAD | Gsplat-mcmc · MipNeRF360 (9 input views) | PSNR: 18.001 -> 18.689 | Improved |
| post-rendering refinement with diffusion prior | Difix3D and HAD · DL3DV (9 sparse views) | PSNR: 21.355 -> 20.673 PSNR; 0.199 -> 0.181 LPIPS | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DepthSplat — DepthSplat: Connecting Gaussian Splatting and Depth (CVPR 2025) · better on 2 of 2 shared results
- GenFusion — GenFusion: Closing the Loop between Reconstruction and Generation via Videos (CVPR 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Hallo3D: Multi-Modal Hallucination Detection and Mitigation for Consistent 3D Content Generation (NeurIPS 2024 · similarity 0.91)
- Mitigating Object Hallucination in Large Vision-Language Models via False Discovery Controlled Visual Data Splitting (NeurIPS 2026 · similarity 0.88)
- Understanding Hallucinations in Diffusion Models through Mode Interpolation (NeurIPS 2024 · similarity 0.88)
- Decompositional Neural Scene Reconstruction with Generative Diffusion Prior (CVPR 2025 · similarity 0.88)
- RODS: Robust Optimization Inspired Diffusion Sampling for Detecting and Reducing Hallucination in Generative Models (NeurIPS 2025 · similarity 0.88)
- Causally-Grounded Dual-Path Attention Intervention for Object Hallucination Mitigation in LVLMs (AAAI 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
