Research map / Video / Video generation
GTA: Advancing Image-to-3D World Generation via Geometry Then Appearance Video Diffusion
Hanxin Zhu, Cong Wang, Peiyan Tu, Jiayi Luo, Tianyu He, Xin Jin, Zhibo 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 |
|---|---|---|---|
| use geometry-then-appearance instead of simultaneous appearance and geometry generation (ablation) | GTA · DL3DV (20 large-viewpoint scenes) | PSNR: 15.03 -> 17.14 | Improved |
| use two-stage geometry-then-appearance instead of a single joint RGB-depth diffusion model (ablation) | GTA · DL3DV eval (55 scenes) | PSNR: 16.62 -> 17.47 | Improved |
| use two-stage geometry-then-appearance instead of a single joint RGB-depth diffusion model (ablation) | GTA · DL3DV eval (55 scenes) | R-err: 0.022 -> 0.013 | Improved |
| use two-stage geometry-then-appearance instead of a single joint RGB-depth diffusion model (ablation) | GTA · DL3DV (20 large-viewpoint scenes) | LPIPS: 0.418 -> 0.367 | Improved |
| add Random Latent Shuffle during training (ablation) | GTA · DL3DV eval (55 scenes) | PSNR: 16.83 -> 17.47 | Improved |
| add Random Latent Shuffle during training (ablation) | GTA · DL3DV eval (55 scenes) | LPIPS: 0.389 -> 0.342 | Improved |
| raise shuffle probability from 0.50 to 1.00 (ablation) | GTA · DL3DV eval (55 scenes) | PSNR: 17.47 -> 17.22 | Hurt |
| raise shuffle probability from 0.00 to 0.50 (ablation) | GTA · DL3DV eval (55 scenes) | LPIPS: 0.389 -> 0.342 | Improved |
| shuffle noisy and target latents but not the conditioning latent (inconsistent shuffle) versus no shuffle (ablation) | GTA · DL3DV eval (55 scenes) | PSNR: 16.83 -> 16.35 | Hurt |
| permute conditioning latents consistently with noisy and target latents versus inconsistent shuffle (ablation) | GTA · DL3DV eval (55 scenes) | PSNR: 16.35 -> 17.47 | Improved |
| add test-time scaling with masked-warping (ablation) | GTA · DL3DV eval (55 scenes) | FID: 38.32 -> 34.54 | Improved |
| add test-time scaling with masked-warping (ablation) | GTA · DL3DV eval (55 scenes) | PSNR: 17.47 -> 17.35 | Hurt |
Closest papers by meaning
- DreamWorld: Geometry-Grounded Video Diffusion for 3D-Consistent World Modeling (ECCV 2026 · similarity 0.90)
- V3D: Video Diffusion Models are Effective 3D Generators (TPAMI 2025 · similarity 0.90)
- 3D Scene Generation: A Survey (IJCV 2026 · similarity 0.89)
- G4Splat: Geometry-Guided Gaussian Splatting with Generative Prior (ICLR 2026 · similarity 0.89)
- GeoVideo: Introducing Geometric Regularization into Video Generation Model (NeurIPS 2025 · similarity 0.88)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · 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.
