Research map / 3D vision / 3D generation
Vitality-Aware Compression for Efficient Image-to-Shape Diffusion Transformers
Jaeah Lee, Hyunjin Kim, Jaewoong Cho, Gihyun Kwon
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 random layer pruning with vitality-aware pruning (ablation) | Step1X-3D · Step1X-3D ablation | Uni3D-I: 0.0829 -> 0.3584 | Improved |
| replace random layer pruning with vitality-aware pruning (ablation) | Hunyuan3D 2.0 · Hunyuan3D 2.0 ablation | Uni3D-I: 0.1171 -> 0.3576 | Improved |
| replace random layer pruning with vitality-aware pruning (ablation) | Hunyuan3D 2mini · Hunyuan3D 2mini ablation | Uni3D-I: 0.3084 -> 0.3437 | Improved |
| use 8-bit instead of 4-bit quantization on pruned model (ablation) | Hunyuan3D 2.0 · Hunyuan3D 2.0 ablation | Uni3D-I: 0.3134 -> 0.3574 | Improved |
| use adaptive 4/8-bit quantization instead of uniform 4-bit (ablation) | Hunyuan3D 2.0 · Hunyuan3D 2.0 ablation | Uni3D-I: 0.3134 -> 0.3528 | Improved |
| use adaptive quantization instead of uniform 8-bit (ablation) | Hunyuan3D 2.0 · Hunyuan3D 2.0 ablation | size (GB): 1.031 -> 0.909 | Improved |
| add selective fine-tuning of Min-vital layers (ablation) | Hunyuan3D 2.0 · Hunyuan3D 2.0 ablation | Uni3D-I: 0.3528 -> 0.3601 | Improved |
| add selective fine-tuning of Min-vital layers (ablation) | Step1X-3D · Step1X-3D ablation | Uni3D-I: 0.3579 -> 0.3580 | No clear effect |
| add selective fine-tuning of Min-vital layers (ablation) | Hunyuan3D 2mini · Hunyuan3D 2mini ablation | Uni3D-I: 0.3437 -> 0.3608 | Improved |
| add selective fine-tuning of Min-vital layers (ablation) | Hunyuan3D 2.0 · Hunyuan3D 2.0 ablation | V-IoU (%): 68.06 -> 72.04 | Improved |
| add selective fine-tuning of Min-vital layers (ablation) | Step1X-3D · Step1X-3D ablation | V-IoU (%): 61.11 -> 71.12 | Improved |
| use separate double/single-block pruning thresholds rather than a shared one (0.18) (ablation) | Hunyuan3D 2.0 · Hunyuan3D 2.0 qualitative | geometry quality | Improved |
Closest papers by meaning
- Layer- and Timestep-Adaptive Differentiable Token Compression Ratios for Efficient Diffusion Transformers (CVPR 2025 · similarity 0.88)
- GPLD3D: Latent Diffusion of 3D Shape Generative Models by Enforcing Geometric and Physical Priors (CVPR 2024 · similarity 0.87)
- Scaling Diffusion Mamba with Bidirectional SSMs for Efficient 3D Shape Generation (AAAI 2025 · similarity 0.87)
- Direct3D-S2: Gigascale 3D Generation Made Easy with Spatial Sparse Attention (NeurIPS 2025 · similarity 0.87)
- Make-A-Shape: a Ten-Million-scale 3D Shape Model (ICML 2024 · similarity 0.86)
- PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers (NeurIPS 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.
