Research map / 3D vision / NeRF and radiance fields
Multi-view Pyramid Transformer: Look Coarser to See Broader
Gyeongjin Kang, Seungkwon Yang, Seungtae Nam, Younggeun Lee, Jungwoo Kim, Eunbyung Park
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 pyramidal feature aggregation from all stages (ablation) | MVP · DL3DV (256x256 ablation) | PSNR: 21.58 -> 22.79 | Improved |
| use forward fine-to-coarse hierarchy instead of reversed hierarchy (ablation) | MVP · DL3DV (256x256 ablation) | PSNR: 18.95 -> 22.79 | Improved |
| use group-wise attention instead of frame-wise attention in stage 2 (ablation) | MVP · DL3DV (256x256 ablation) | PSNR: 22.53 -> 22.79 | Improved |
| use global attention instead of group-wise attention in stage 2 (ablation) | MVP · DL3DV (256x256 ablation) | PSNR: 22.79 -> 22.94 | Mixed |
| add inter-view hierarchy (local-to-global attention) (ablation) | MVP · DL3DV (256x256 ablation) | PSNR: 22.94 -> 22.79 | Mixed |
| add intra-view token merging hierarchy (ablation) | MVP · DL3DV (256x256 ablation) | LPIPS: 0.249 -> 0.235 | Improved |
| add register tokens (ablation) | MVP · DL3DV (256x256 ablation) | PSNR: 22.52 -> 22.79 | Improved |
| use group size 4 instead of group size 2 (ablation) | MVP · DL3DV (256x256 ablation) | PSNR: 22.94 -> 23.18 | Improved |
| use group size 8 instead of group size 4 (ablation) | MVP · DL3DV (256x256 ablation) | PSNR: 23.18 -> 23.18 | No clear effect |
| use fine patch sizes (4, 8, 16) instead of coarse (8, 16, 32) (ablation) | MVP · RE10K (4 views, 256x256) | PSNR: 30.56 -> 32.12 | Improved |
| prune Gaussians with opacity below 0.01 at render time | MVP · DL3DV | PSNR: -0.02 to -0.15 PSNR with 79.7% to 89.1% pruned | 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.
- Long-LRM — Long-LRM: Long-sequence Large Reconstruction Model for Wide-coverage Gaussian Splats (ICCV 2025) · better on 4 of 4 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
- Déjà View: Looping Transformers for Multi-View 3D Reconstruction (NeurIPS 2026 · similarity 0.89)
- Understanding Multi-View Transformers (NeurIPS 2026 · similarity 0.89)
- Fast3R: Towards 3D Reconstruction of 1000+ Images in One Forward Pass (CVPR 2025 · similarity 0.88)
- Inverted Pyramid Multi-task Transformer for Dense Scene Understanding (ECCV 2022 · similarity 0.88)
- LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias (ICLR 2025 · similarity 0.87)
- TurboVGGT: Fast Visual Geometry Reconstruction with Adaptive Alternating Attention (NeurIPS 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.
