Research map / Image generation / Fast diffusion sampling
Foresight Diffusion: Improving Sampling Consistency in Predictive Diffusion Models
Yu Zhang, Xingzhuo Guo, Haoran Xu, Jialong Wu, Mingsheng Long
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 two-stage deterministic pretraining of the predictive stream (ForeDiff vs ForeDiff-zero) (ablation) | ForeDiff · RoboNet | STD of LPIPS: 0.66 -> 0.35 | Improved |
| add two-stage deterministic pretraining of the predictive stream (ForeDiff vs ForeDiff-zero) (ablation) | ForeDiff · RT-1 | STD of LPIPS: 0.50 -> 0.17 | Improved |
| add two-stage deterministic pretraining of the predictive stream (ForeDiff vs ForeDiff-zero) (ablation) | ForeDiff · HeterNS | relative L2: 0.83 -> 0.18 | Improved |
| add architectural decoupling alone (ForeDiff-zero) to vanilla diffusion (ablation) | Vanilla Diffusion · RoboNet | LPIPS (x100): 5.65 -> 5.54 | No clear effect |
| add architectural decoupling alone (ForeDiff-zero) to vanilla diffusion (ablation) | Vanilla Diffusion · HeterNS | relative L2: 1.50 -> 0.83 | Improved |
| add predictive ViT stream and pretraining (full ForeDiff) to vanilla diffusion (ablation) | Vanilla Diffusion · RoboNet | FVD: 53.8 -> 51.5 | Improved |
| add predictive ViT stream and pretraining (full ForeDiff) to vanilla diffusion (ablation) | Vanilla Diffusion · RT-1 | FVD: 11.7 -> 12.0 | Hurt |
| condition on internal predictive features instead of PredHead outputs (ablation) | ForeDiff · RoboNet | FVD: 53.7 -> 51.5 | Improved |
| increase predictive ViT blocks from M=0 to M=6 (ablation) | ForeDiff · RoboNet | LPIPS (x100): 5.65 -> 5.25 | Improved |
| increase predictive ViT blocks from M=6 to M=12 (ablation) | ForeDiff · RoboNet | FVD: 51.5 -> 52.1 | Hurt |
| add classifier-free guidance at inference (ablation) | Vanilla Diffusion · RoboNet | LPIPS (x100): 5.65 -> 5.27 | Improved |
| add classifier-free guidance at inference (ablation) | ForeDiff · RoboNet | LPIPS (x100): 5.25 -> 5.05 | Improved |
Closest papers by meaning
- DiRecT: Safe Diffusion-Based Planning via Receding-Horizon Denoising (NeurIPS 2026 · similarity 0.89)
- Prediction with Action: Visual Policy Learning via Joint Denoising Process (NeurIPS 2024 · similarity 0.88)
- Coarse-to-Fine Compositional Diffusion for Long-Horizon Planning (NeurIPS 2026 · similarity 0.88)
- Dynamical Diffusion: Learning Temporal Dynamics with Diffusion Models (ICLR 2025 · similarity 0.87)
- Interpreting and Improving Diffusion Models from an Optimization Perspective (ICML 2024 · similarity 0.87)
- Learning to Sample From Diffusion Models via Inverse Reinforcement Learning (NeurIPS 2026 · 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.
