Research map / 3D vision / 3D generation
Scaling Sequence-to-Sequence Generative Neural Rendering
Shikun Liu, Kam Woh Ng, Wonbong Jang, Jiadong Guo, Junlin Han, Haozhe Liu, Yiannis Douratsos, Juan Camilo Perez, Zijian Zhou, Khanh Chi Phung, Tao Xiang, Juan-Manuel Perez-Rua
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 |
|---|---|---|---|
| shift timestep sampler toward noise (Mode, scale 0.8, shift 3) vs unshifted Mode (ablation) | Kaleido-Small · Objaverse PSNR 1->5 | PSNR: 17.39 -> 18.19 | Improved |
| use LinearQuadratic inference schedule vs trailing schedule (ablation) | Kaleido-Small · Objaverse PSNR 1->5 | PSNR: 17.95 -> 18.09 | Improved |
| add video pre-training (200K steps) (ablation) | Kaleido-Small · Objaverse PSNR 5->5 | PSNR: 23.95 -> 24.60 | Improved |
| add one register token to attention (ablation) | Kaleido-Small · massive activation magnitude | top activation: 15192 -> 397.75 | Improved |
| use 8 registers instead of 1 (ablation) | Kaleido-Small · Objaverse PSNR 1->5 | PSNR: 15.93 -> 15.26 | Hurt |
| exponential view sampling with masking vs uniform with masking (ablation) | Kaleido-Small · Objaverse PSNR 1->5 | PSNR: 14.51 -> 15.13 | Improved |
| exponential view sampling vs uniform without masking (ablation) | Kaleido-Small · Objaverse PSNR 5->1 | PSNR: 22.22 -> 21.41 | Mixed |
| increase temporal window attention size K from 4 to 8 (ablation) | Kaleido-Small · uCO3D PSNR 5->5 | PSNR: 18.45 -> 19.15 | Improved |
| temporal window attention K=4 instead of full cross-view attention (ablation) | Kaleido-Small · uCO3D PSNR 5->5 | PSNR: 18.28 -> 18.45 | No clear effect |
| add DINOv2-B auxiliary features (ablation) | Kaleido-Small · Objaverse PSNR 1->5 | PSNR: 15.67 -> 15.86 | Improved |
| use DINOv2-L instead of DINOv2-B (ablation) | Kaleido-Small · uCO3D PSNR 5->5 | PSNR: 18.81 -> 18.75 | No clear effect |
| remove timestep conditioning (ablation) | Kaleido-Small · Objaverse PSNR 5->1 | PSNR: 22.28 -> 20.85 | Hurt |
Closest papers by meaning
- UMAMI: Unifying Masked Autoregressive Models and Deterministic Rendering for View Synthesis (NeurIPS 2025 · similarity 0.89)
- MVGenMaster: Scaling Multi-View Generation from Any Image via 3D Priors Enhanced Diffusion Model (CVPR 2025 · similarity 0.88)
- DreamScene4D: Dynamic Multi-Object Scene Generation from Monocular Videos (NeurIPS 2024 · similarity 0.87)
- 4Real: Towards Photorealistic 4D Scene Generation via Video Diffusion Models (NeurIPS 2024 · similarity 0.87)
- Coarse-to-Real: Generative Rendering for Populated Dynamic Scenes (NeurIPS 2026 · similarity 0.87)
- Efficient-3Dim: Learning a Generalizable Single-image Novel-view Synthesizer in One Day (ICLR 2024 · 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.
