Research map / 3D vision / Gaussian splatting
UniQueR: Unified Query-based Feedforward 3D Reconstruction
Chensheng Peng, Quentin HERAU, Jiezhi Yang, Yichen Xie, Yihan Hu, Wenzhao Zheng, Matthew Strong, Masayoshi TOMIZUKA, Wei Zhan
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 |
|---|---|---|---|
| Add rendered-depth supervision (ablation) | UniQueR · Mip-NeRF 360 (3 views) | PSNR/LPIPS: PSNR 19.96 -> 20.23; LPIPS 0.234 -> 0.182; SSIM 0.718 -> 0.713 | Mixed |
| Hybrid query init (half point-map, half uniform) vs point-map only (ablation) | UniQueR · Mip-NeRF 360 (3 views) | PSNR: 18.58 -> 20.23 | Improved |
| Hybrid init vs random-only init (ablation) | UniQueR · Mip-NeRF 360 (3 views) | PSNR: 12.11 -> 20.23 | Improved |
| Increase number of queries (ablation) | UniQueR · Mip-NeRF 360 | PSNR | Improved |
| Increase Gaussians per query 16 -> 32 -> 64 (ablation) | UniQueR · Mip-NeRF 360 | PSNR | Improved |
| Increase transformer model size (ablation) | UniQueR · Mip-NeRF 360 | PSNR | Improved |
| Replace AnySplat with UniQueR (sparse view) | feedforward NVS · Mip-NeRF 360 (3 views) | PSNR: 20.08 -> 22.70 | Improved |
| Replace AnySplat with UniQueR (sparse view) | feedforward NVS · VR-NeRF (3 views) | SSIM/LPIPS: SSIM 0.745 -> 0.708; LPIPS 0.313 -> 0.446 | Hurt |
| Replace AnySplat with UniQueR (dense feedforward only) | feedforward NVS · Mip-NeRF 360 (32 views) | PSNR: 22.32 -> 21.51 | Hurt |
| Use UniQueR as init for 3DGS instead of AnySplat | 3DGS · VR-NeRF (32 views) | PSNR: 21.90 -> 27.03 | Improved |
| Replace AnySplat with UniQueR (efficiency) | feedforward NVS · 32 views, 448^2 | depth Abs Rel / time: 0.062 -> 0.038; 4.63s -> 1.97s | Improved |
| Replace Pi3 with UniQueR for pose | camera pose · RealEstate10K | AUC@30: 85.90 -> 83.69 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FLARE — FLARE: Feed-forward Geometry, Appearance and Camera Estimation from Uncalibrated Sparse Views (CVPR 2025) · better on 1 of 1 shared result
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
- Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Image (NeurIPS 2024 · similarity 0.88)
- SPAR3D: Stable Point-Aware Reconstruction of 3D Objects from Single Images (CVPR 2025 · similarity 0.88)
- Fin3R: Fine-tuning Feed-forward 3D Reconstruction Models via Monocular Knowledge Distillation (NeurIPS 2025 · similarity 0.87)
- SPARS3R: Semantic Prior Alignment and Regularization for Sparse 3D Reconstruction (CVPR 2025 · similarity 0.87)
- CompactSplat: Spatially Adaptive Gaussian Distribution for Feedforward Scene Reconstruction (NeurIPS 2026 · similarity 0.86)
- Test3R: Learning to Reconstruct 3D at Test Time (NeurIPS 2025 · similarity 0.86)
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.
