Research map / Image generation / Fast diffusion sampling
Test-Time Anchoring for Discrete Diffusion Posterior Sampling
Litu Rout, Andreas Lugmayr, Yasamin Jafarian, Srivatsan Varadharajan, Constantine Caramanis, Sanjay Shakkottai, Ira Kemelmacher-Shlizerman
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 |
|---|---|---|---|
| use APS instead of G2D2 | discrete diffusion sampler · FFHQ SR (4x) | LPIPS: APS reduces LPIPS by 13.65% | Improved |
| use APS instead of G2D2 | discrete diffusion sampler · FFHQ SR (4x) | PSNR: improves PSNR by 2.11% | Improved |
| use APS instead of G2D2 | discrete diffusion sampler · FFHQ Gaussian deblur | LPIPS: lowers LPIPS by 3.83% | Improved |
| use APS instead of G2D2 | discrete diffusion sampler · FFHQ Gaussian deblur | PSNR: raises PSNR by 5.88% | Improved |
| use large-resolution variant APS-L (1024x1024 upsampled) instead of G2D2 | APS · FFHQ | LPIPS: up to a 31.36% gain over G2D2 | Improved |
| use APS instead of G2D2 | discrete diffusion sampler · ImageNet SR (4x) | LPIPS: reduces LPIPS by 7.16% | Improved |
| use APS instead of G2D2 | discrete diffusion sampler · ImageNet SR (4x) | PSNR: improves PSNR by 4.74% | Improved |
| use APS instead of G2D2 | discrete diffusion sampler · ImageNet Gaussian deblur | PSNR: improves PSNR by 8.81% | Improved |
| use APS-L instead of G2D2 | APS · ImageNet | LPIPS: improves LPIPS by up to 35.82% | Improved |
| add APS guidance to MMaDA | MMaDA · Reference-based stylization | ImageReward: 0.48 -> 0.63 | Improved |
| add APS guidance to MMaDA | MMaDA · Reference-based stylization | DINO: 0.32 -> 0.41 | Improved |
| add APS with Qwen3-0.6B reward model to Dream-7B-Instruct | Dream-7B-Instruct · Question answering (20 questions) | Oracle judge score: 3.7523 -> 4.5775 | Improved |
Closest papers by meaning
- Anchored Diffusion Language Model (NeurIPS 2025 · similarity 0.90)
- Encoder-Decoder Diffusion Language Models for Efficient Training and Inference (NeurIPS 2025 · similarity 0.89)
- Zigzag Diffusion Sampling: Diffusion Models Can Self-Improve via Self-Reflection (ICLR 2025 · similarity 0.88)
- Fast Sampling via Discrete Non-Markov Diffusion Models with Predetermined Transition Time (NeurIPS 2024 · similarity 0.88)
- Fast constrained sampling in pre-trained diffusion models (NeurIPS 2025 · similarity 0.88)
- On Discrete Prompt Optimization for Diffusion Models (ICML 2024 · 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.
