Research map / 3D vision / Gaussian splatting
Learning Compact 3D Representations from Feed-Forward Novel View Synthesis
Honggyu An, Jaewoo Jung, Mungyeom Kim, Chaehyun Kim, Minkyeong Jeon, Jisang Han, Kazumi Fukuda, Takuya Narihira, Hyunah Ko, Junsu Kim, Sunghwan Hong, Yuki Mitsufuji and 1 more
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 progressive low-pass filter during training (ablation) | C3G-G · RealEstate10K | PSNR: N/A -> 22.387 | Improved |
| unfreeze the visual encoder (ablation) | C3G-G · RealEstate10K | PSNR: 18.441 -> 22.387 | Improved |
| increase learnable query tokens from 2048 to 4096 (ablation) | C3G-G · RealEstate10K | PSNR: 20.625 -> 19.012 | Hurt |
| increase learnable query tokens from 256 to 2048 (ablation) | C3G-G · RealEstate10K | PSNR: 19.713 -> 20.625 | Improved |
| use VGGT instead of DINOv3 as visual encoder (ablation) | C3G-G · RealEstate10K | PSNR: 20.292 -> 22.387 | Improved |
| use C3G-F feature decoder instead of none (ablation) | C3G-F · ScanNet (LSeg) | mIoU: 0.193 -> 0.513 | Improved |
| add autoencoder feature compression to C3G-F (ablation) | C3G-F · ScanNet (LSeg) | mIoU: 0.513 -> 0.512 | No clear effect |
| add C3G feature lifting on top of VGGT-Tracking features | VGGT-Tracking · ScanNet correspondence | average PCK@10px: 34.0 -> 68.1 | Improved |
| add AnyUp upsampler to DINOv3-Large features | DINOv3-Large · ScanNet correspondence | average PCK@10px: 37.7 -> 31.8 | Hurt |
| add AnyUp upsampler to DINOv2-Base features | DINOv2-Base · ScanNet correspondence | average PCK@10px: 23.8 -> 23.9 | No clear effect |
| use more input views with per-pixel Gaussians (VGGT+NoPo) | VGGT+NoPo · RealEstate10K | PSNR: 21.262 -> 21.190 | Hurt |
| add test-time optimization | C3G · RealEstate10K 36 views | PSNR: 23.812 -> 30.250 | Improved |
Closest papers by meaning
- FreeSplat: Generalizable 3D Gaussian Splatting Towards Free View Synthesis of Indoor Scenes (NeurIPS 2024 · similarity 0.90)
- No Pose, No Problem: Surprisingly Simple 3D Gaussian Splats from Sparse Unposed Images (ICLR 2025 · similarity 0.90)
- FHGS: Feature-Homogenized Gaussian Splatting (NeurIPS 2025 · similarity 0.90)
- PF3plat: Pose-Free Feed-Forward 3D Gaussian Splatting for Novel View Synthesis (ICML 2025 · similarity 0.89)
- CompactSplat: Spatially Adaptive Gaussian Distribution for Feedforward Scene Reconstruction (NeurIPS 2026 · similarity 0.89)
- Generative Densification: Learning to Densify Gaussians for High-Fidelity Generalizable 3D Reconstruction (CVPR 2025 · similarity 0.89)
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.
