Research map / 3D vision / 3D generation
Compress3D: a Compressed Latent Space for 3D Generation from a Single Image
Bowen Zhang, Tianyu Yang, Yu Li, Lei Zhang, Xi Zhao
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 3D-aware cross-attention querying a feature volume in the triplane autoencoder (ablation) | Triplane autoencoder · Autoencoder reconstruction | L_rgb x10^3: 3.798 -> 2.485 | Improved |
| add 3D-aware cross-attention (training cost) (ablation) | Triplane autoencoder · Autoencoder training | seconds per step: 0.789 -> 0.824 | Hurt |
| use down-sampled 32 feature volume instead of 128 in cross-attention (ablation) | Triplane autoencoder · Autoencoder reconstruction | L_rgb x10^3: 2.551 -> 2.485 | Improved |
| use 32 volume resolution instead of 128 for training speed (ablation) | Triplane autoencoder · Autoencoder training | seconds per step: 2.295 -> 0.824 | Improved |
| add diffusion prior model producing shape embedding as extra condition (ablation) | Compress3D · Single-image 3D generation | FID: 66.46 -> 53.21 | Improved |
| add diffusion prior model producing shape embedding as extra condition (ablation) | Compress3D · Single-image 3D generation | CLIP similarity: 0.745 -> 0.776 | Improved |
| raise image guidance scale s_p from 1.0 to 5.0 at s_s=1.0 (ablation) | Compress3D · Single-image 3D generation | FID: 65.18 -> 53.21 | Improved |
| increase guidance scales to 10.0 for both (ablation) | Compress3D · Single-image 3D generation | FID: 53.21 -> 54.91 | Hurt |
| use highly compressed triplane latent instead of Shap-E latent | Compress3D · Single-image 3D generation | FID: 146.14 -> 53.21 | Improved |
Closest papers by meaning
- LaGeM: A Large Geometry Model for 3D Representation Learning and Diffusion (ICLR 2025 · similarity 0.89)
- Direct3D-S2: Gigascale 3D Generation Made Easy with Spatial Sparse Attention (NeurIPS 2025 · similarity 0.89)
- Enhancing High-Resolution 3D Generation through Pixel-wise Gradient Clipping (ICLR 2024 · similarity 0.88)
- EdgeRunner: Auto-regressive Auto-encoder for Artistic Mesh Generation (ICLR 2025 · similarity 0.88)
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 2025 · similarity 0.87)
- Structured 3D Latents for Scalable and Versatile 3D Generation (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.
