Research map / Image generation / Autoregressive and latent generation
RePack then Refine: Efficient Diffusion Transformers with Vision Foundation Models
Guanfang Dong, Luke Schultz, Negar Hassanpour, Chao Gao
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 Latent-Guided Refiner (ablation) | RePack-DiT-XL/1 · ImageNet 256x256 (CFG) | gFID: 1.82 -> 1.65 | Improved |
| add Latent-Guided Refiner (ablation) | RePack · ImageNet reconstruction | rFID: 0.83 -> 0.69 | Improved |
| add Refiner (ablation) | RePack-DiT-B/1 · ImageNet (64 epochs) | gFID: 4.12 -> 3.99 | Improved |
| add Refiner (ablation) | RePack-DiT-XL/2 · ImageNet (64 epochs) | gFID: 3.66 -> 3.20 | Improved |
| add Refiner with fewer sampling steps (25 steps) (ablation) | RePack-DiT-XL/1 · ImageNet (64 epochs) | gFID: 2.09 -> 2.04 | Improved |
| add Refiner without CFG (ablation) | RePack · ImageNet 256x256 (no CFG) | gFID: 3.81 -> 3.79 | No clear effect |
| apply Refiner to VA-VAE or SD-VAE (ablation) | VA-VAE, SD-VAE · ImageNet | gFID | Hurt |
| use bias-free linear projection instead of linear with bias (ablation) | RePack · ImageNet (DiT-B/1, 64 epochs) | gFID: 5.08 -> 4.12 | Improved |
| use learnable linear projection instead of PCA (ablation) | RePack · ImageNet (DiT-B/1, 64 epochs) | gFID: 7.56 -> 4.12 | Improved |
| reduce latent dimension from 32 to 16 (ablation) | RePack · ImageNet (DiT-B/1, 64 epochs) | gFID: 4.12 -> 3.61 | Improved |
| increase latent dimension from 32 to raw 768 (ablation) | RePack · ImageNet (DiT-B/1, 64 epochs) | gFID: 4.12 -> 66.43 | Hurt |
| switch VFM from DINOv3-B to DINOv2-B (ablation) | RePack · ImageNet (DiT-B/1, 64 epochs) | gFID: 4.12 -> 3.84 | Improved |
Closest papers by meaning
- RepLDM: Reprogramming Pretrained Latent Diffusion Models for High-Quality, High-Efficiency, High-Resolution Image Generation (NeurIPS 2025 · similarity 0.90)
- Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models (CVPR 2025 · similarity 0.89)
- Scaling Diffusion Transformers Efficiently via $\mu$P (NeurIPS 2025 · similarity 0.89)
- REPA Works Until It Doesn’t: Early-Stopped, Holistic Alignment Supercharges Diffusion Training (NeurIPS 2025 · similarity 0.89)
- Learning Stackable and Skippable LEGO Bricks for Efficient, Reconfigurable, and Variable-Resolution Diffusion Modeling (ICLR 2024 · similarity 0.89)
- BlockDance: Reuse Structurally Similar Spatio-Temporal Features to Accelerate Diffusion Transformers (CVPR 2025 · similarity 0.88)
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.
