Research map / 3D vision / 3D generation
EarthCrafter: Scalable 3D Earth Generation via Dual-Sparse Latent Diffusion
Shang Liu, Chenjie Cao, Chaohui Yu, Wen Qian, Jing Wang, Fan Wang
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 low-channel high-resolution FLUX-VAE features instead of coarse large-channel DINO features (ablation) | TexVAE · TexVAE ablation (NY subset) | PSNR: 17.76 -> 18.90 | Improved |
| increase voxel resolution from L=200 to L=360 (ablation) | TexVAE · TexVAE ablation (NY subset) | PSNR: 17.36 -> 18.90 | Improved |
| add multi-scale FLUX features plus RGB and normal features (ablation) | TexVAE · TexVAE ablation (NY subset) | PSNR: 18.90 -> 19.49 | Improved |
| spatially compress TexVAE with a UNet instead of tube network (ablation) | TexVAE · TexVAE ablation (NY subset) | PSNR: 19.49 -> 16.22 | Hurt |
| add full attention at lowest resolution layer (ablation) | StructVAE · StructVAE accuracy | voxel accuracy: 95.3 -> 94.9 (NY ablation data); 96.6 -> 97.1 (global data) | Mixed |
| replace convolution-based XCube with transformer-based StructVAE (ablation) | StructVAE · StructVAE (ablation NY data) | voxel accuracy: 94.3 -> 95.3 | Improved |
| use two-stage coarse-to-fine structure flow matching instead of one-stage dense (ablation) | StructFM · StructFM (image condition, mIoU at L/8) | mIoU 3: 82.8 -> 86.1 | Improved |
| use two-stage coarse-to-fine structure flow matching instead of one-stage dense (ablation) | StructFM · StructFM (image condition, mIoU at L level) | mIoU 0: 18.9 -> 25.4 | Improved |
Closest papers by meaning
- GeoCore-9B: Towards Geo-Aware Generative Foundation Models in Earth Observation (NeurIPS 2026 · similarity 0.88)
- Around the World in 80 Timesteps: A Generative Approach to Global Visual Geolocation (CVPR 2025 · similarity 0.87)
- Terrain Diffusion Network: Climatic-Aware Terrain Generation with Geological Sketch Guidance (AAAI 2024 · similarity 0.87)
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 2025 · similarity 0.86)
- No Location Left Behind: Measuring and Improving the Fairness of Implicit Representations for Earth Data (ICLR 2025 · similarity 0.86)
- AerialMegaDepth: Learning Aerial-Ground Reconstruction and View Synthesis (CVPR 2025 · similarity 0.86)
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.
