Research map / Image generation / Fast diffusion sampling
Transition Models: Rethinking the Generative Learning Objective
Zidong Wang, Yiyuan Zhang, Xiaoyu Yue, Xiangyu Yue, Yangguang Li, Wanli Ouyang, Lei Bai
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 SiT objective with TiM transition objective (ablation) | SiT-B/4 · ImageNet-256 | FID (NFE=1): 309.5 -> 49.91 | Improved |
| replace SiT objective with TiM transition objective (ablation) | SiT-B/4 · ImageNet-256 | FID (NFE=50): 20.35 -> 17.99 | Improved |
| use finite-difference DDE instead of JVP (ablation) | TiM-B/4 · ImageNet-256 | FID (NFE=1): 49.75 -> 49.91 | No clear effect |
| use finite-difference DDE instead of JVP (ablation) | TiM-B/4 · training cost | throughput: 1.80 -> 2.40 (/s) | Improved |
| add decoupled time embedding (ablation) | TiM-B/4 · ImageNet-256 | FID (NFE=1): 56.22 -> 49.91 | Improved |
| add interval-aware attention (ablation) | TiM-B/4 · ImageNet-256 | FID (NFE=1): 56.22 -> 48.38 | Improved |
| combine decoupled time embedding and interval-aware attention (ablation) | TiM-B/4 · ImageNet-256 | FID (NFE=50): 20.37 -> 17.43 | Improved |
| add interval-aware attention on top of decoupled time embedding (ablation) | TiM-B/4 · ImageNet-256 | FID (NFE=8): 26.09 -> 26.10 | No clear effect |
| add time-weighting favoring short intervals (ablation) | TiM-B/4 · ImageNet-256 | FID (NFE=1): 48.30 -> 47.46 | Improved |
| increase sampling NFE from 1 to 128 (ablation) | TiM · GenEval | overall score: 0.67 -> 0.83 | Improved |
| increase sampling NFE from 1 to 128 (ablation) | FLUX.1-Schnell · GenEval | overall score: 0.68 -> 0.58 | Hurt |
| increase NFE from 8 to 32 (ablation) | TiM · MJHQ30K | FID: 5.28 -> 5.65 | Mixed |
Closest papers by meaning
- One-Step Offline Distillation of Diffusion-based Models via Koopman Modeling (NeurIPS 2025 · similarity 0.90)
- Generative Modeling with Phase Stochastic Bridge (ICLR 2024 · similarity 0.89)
- A solvable model of learning generative diffusion: theory and insights (NeurIPS 2025 · similarity 0.88)
- Scaling Inference Time Compute for Diffusion Models (CVPR 2025 · similarity 0.88)
- Action-Minimization Meets Generative Modeling: Efficient Transition Path Sampling with the Onsager-Machlup Functional (ICML 2025 · similarity 0.88)
- CCS: Controllable and Constrained Sampling with Diffusion Models via Initial Noise Perturbation (NeurIPS 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.
