Research map / 3D vision / 3D generation
Large Scene Generation with Cube-Absorb Discrete Diffusion
Qianjiang Hu, Wei Hu
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 cube-absorb diffusion (CADD) instead of uniform diffusion (ablation) | Conv-based hierarchical model · CarlaSC (component ablation, unconditional) | mIoU (SparseUNet): 67.59 -> 78.34 | Improved |
| add SCDT instead of convolution (ablation) | CADD · CarlaSC (component ablation, unconditional) | mIoU (PointNet++): 49.73 -> 55.08 | Improved |
| add SCDT instead of convolution (ablation) | CADD · CarlaSC (component ablation, unconditional) | mIoU (SparseUNet): 78.34 -> 78.13 | No clear effect |
| use cube-absorb instead of uniform diffusion (ablation) | coarse-to-fine refinement · CarlaSC (diffusion process ablation, 128x128x8 to 256x256x16) | mIoU: 92.69 -> 95.18 | Improved |
| use cube-absorb instead of absorb diffusion (ablation) | coarse-to-fine refinement · CarlaSC (diffusion process ablation, 128x128x8 to 256x256x16) | mIoU: 94.02 -> 95.18 | Improved |
| use cube-absorb instead of uniform diffusion (ablation) | coarse-to-fine refinement · CarlaSC (diffusion process ablation, 128x128x8 to 256x256x16) | F3D: 0.229 -> 0.175 | Improved |
| use SCDT instead of Cube-DiT (empty cube embedding) (ablation) | sparse DiT · CarlaSC (architecture ablation, 128x128x8 to 256x256x16) | Memory usage: 17.09G -> 6.38G | Improved |
| use SCDT instead of SparseConv (ablation) | sparse generator · CarlaSC (architecture ablation, 128x128x8 to 256x256x16) | mIoU: 94.37 -> 95.18 | Improved |
| use sparse cube local attention instead of original global DiT (ablation) | DiT · CarlaSC (architecture ablation, 128x128x8 to 256x256x16) | feasibility: OOM -> 6.38G | Improved |
| use a 3-level hierarchy instead of a single level (ablation) | CADD · CarlaSC (hierarchy ablation, 256x256x16 single level) | mIoU (SparseUNet): 45.39 -> 78.13 | Improved |
| add a fourth level starting from 32x32x2 (ablation) | CADD hierarchy · CarlaSC (hierarchy ablation, 256x256x16 single level) | mIoU (SparseUNet): 78.13 -> 81.07 | No clear effect |
| use CADD instead of PDD | hierarchical diffusion · CarlaSC (unconditional) | mIoU (SparseUNet): 68.02 -> 78.13 | Improved |
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.
- XCube — XCube: Large-Scale 3D Generative Modeling using Sparse Voxel Hierarchies (CVPR 2024) · 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
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 2025 · similarity 0.88)
- Denoising Diffusion via Image-Based Rendering (ICLR 2024 · similarity 0.88)
- XCube: Large-Scale 3D Generative Modeling using Sparse Voxel Hierarchies (CVPR 2024 · similarity 0.87)
- CubeDiff: Repurposing Diffusion-Based Image Models for Panorama Generation (ICLR 2025 · similarity 0.87)
- One-2-3-45++: Fast Single Image to 3D Objects with Consistent Multi-View Generation and 3D Diffusion (CVPR 2024 · similarity 0.87)
- Wonderland: Navigating 3D Scenes from a Single Image (CVPR 2025 · 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.
