Research map / 3D vision / 3D generation
AssetFormer: Modular 3D Assets Generation with Autoregressive Transformer
Lingting Zhu, Shengju Qian, Haidi Fan, Jiayu Dong, Zhenchao Jin, SiweiZhou, Dong Gen, Xin Wang, Lequan Yu
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 |
|---|---|---|---|
| replace PCG with AssetFormer | AssetFormer · modular building assets | FID: 108.476 -> 55.186 | Improved |
| use top-k sampling instead of greedy search (ablation) | AssetFormer · modular building assets | FID: 63.351 -> 55.186 | Improved |
| use top-k sampling instead of beam search (ablation) | AssetFormer · modular building assets | FID: 63.333 -> 55.186 | Improved |
| re-order tokens with DFS instead of raw order (ablation) | AssetFormer · modular building assets | FID: 65.215 -> 55.186 | Improved |
| re-order tokens with DFS instead of BFS (ablation) | AssetFormer · modular building assets | FID: 61.620 -> 55.186 | Improved |
| use randomized token order (RAR) instead of DFS (ablation) | AssetFormer · modular building assets | FID: 55.186 -> 83.561 | Hurt |
| add real collected data to synthesized data (ablation) | AssetFormer · modular building assets | FID: 113.560 -> 55.186 | Improved |
| add synthesized PCG data to collected data (ablation) | AssetFormer · modular building assets | FID: 63.381 -> 55.186 | Improved |
| SlowFast decoding with small draft model vs target alone | AssetFormer-B · modular building assets | speed (token/s): 80.62 -> 119.02 | Improved |
| SlowFast decoding vs target alone on quality | AssetFormer-B · modular building assets | FID: 55.186 -> 55.831 | No clear effect |
| choice of sampling method on CLIP score (ablation) | AssetFormer · modular building assets | CLIP: 0.319 / 0.321 / 0.320 | No clear effect |
Closest papers by meaning
- Structured 3D Latents for Scalable and Versatile 3D Generation (CVPR 2025 · similarity 0.85)
- Meta 3D AssetGen: Text-to-Mesh Generation with High-Quality Geometry, Texture, and PBR Materials (NeurIPS 2024 · similarity 0.85)
- AutoPartGen: Autoregressive 3D Part Generation and Discovery (NeurIPS 2025 · similarity 0.84)
- DecompDreamer: A Composition-Aware Curriculum for Structured 3D Asset Generation (NeurIPS 2026 · similarity 0.84)
- 3DTopia-XL: Scaling High-quality 3D Asset Generation via Primitive Diffusion (CVPR 2025 · similarity 0.84)
- ViewDiff: 3D-Consistent Image Generation with Text-to-Image Models (CVPR 2024 · similarity 0.83)
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.
