Research map / 3D vision / 3D generation
The Less You Depend, The More You Learn: Synthesizing Novel Views from Sparse, Unposed Images without Any 3D Knowledge
Haoru Wang, Kai Ye, Minghan Qin, Yangyan Li, Wenzheng Chen, Baoquan Chen
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 |
|---|---|---|---|
| reduce explicit 3D scene structure (LVSM vs MVSplat) | feed-forward NVS · RealEstate10K scalability | avg PSNR gain per 4x data: 0.39 -> 0.64 | Improved |
| remove input pose requirement (PT-LVSM vs LVSM) | feed-forward NVS · RealEstate10K scalability | avg PSNR gain per 4x data: 0.64 -> 1.72 | Improved |
| remove target pose requirement (UP-LVSM vs PT-LVSM) | feed-forward NVS · RealEstate10K scalability | avg PSNR gain per 4x data: 1.72 -> 2.63 | Improved |
| explicit 3D structure at low data (NoPoSplat/MVSplat vs implicit) | feed-forward NVS · RealEstate10K little subset (1K scenes) | PSNR: MVSplat 25.24 vs PT-LVSM 20.80 | Mixed |
| use latent Plucker learner rather than SfM target poses (ablation) | UP-LVSM · RealEstate10K | PSNR: 26.00 -> 28.82 | Improved |
| use latent Plucker learner rather than key-value pose estimator (ablation) | UP-LVSM · RealEstate10K | PSNR: 20.92 -> 28.82 | Improved |
| train UP-LVSM without any poses vs pose-dependent LVSM | UP-LVSM · RealEstate10K full eval | PSNR: 27.60 -> 28.82 | Improved |
| less dependence at small data on ACID: UP-LVSM at 0.8K scenes | UP-LVSM · ACID 0.8K scenes | PSNR: LVSM 23.43 vs UP-LVSM 15.92 | Hurt |
| train on RealEstate10K and test zero-shot on ACID | UP-LVSM · ACID | PSNR: 27.21 -> 27.33 | Improved |
| pose-free UP-LVSM on Objaverse at 128K objects vs LVSM | UP-LVSM · Objaverse 128K objects | PSNR: 30.22 vs 26.12 | Hurt |
| fine-tune with linear SE(3)-to-latent mapper on 1202 posed scenes | UP-LVSM · RealEstate10K | PSNR | No clear effect |
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.
- MVSplat — MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images (ECCV 2024) · better on 2 of 3 shared results
- PixelSplat — pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction (CVPR 2024) · 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
- LEAP: Liberate Sparse-View 3D Modeling from Camera Poses (ICLR 2024 · similarity 0.89)
- MVSplat360: Feed-Forward 360 Scene Synthesis from Sparse Views (NeurIPS 2024 · similarity 0.89)
- Sparse-view Pose Estimation and Reconstruction via Analysis by Generative Synthesis (NeurIPS 2024 · similarity 0.89)
- Novel View Synthesis with Pixel-Space Diffusion Models (CVPR 2025 · similarity 0.88)
- Efficient-3Dim: Learning a Generalizable Single-image Novel-view Synthesizer in One Day (ICLR 2024 · similarity 0.88)
- The More You See in 2D the More You Perceive in 3D (CVPR 2024 · similarity 0.88)
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.
