Research map / Image generation / Fast diffusion sampling
Test-Time Iterative Error Correction for Efficient Diffusion Models
Yunshan Zhong, Weiqi Yan, Yuxin Zhang
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 IEC to W4A8 quantized model | timestep-wise quantization · CIFAR-10 | FID: 6.82 -> 5.96 | Improved |
| add IEC to W4A8 quantized model | timestep-wise quantization · LSUN-Churches | FID: 6.27 -> 6.10 | Improved |
| add IEC to DeepCache N=15 | DeepCache · CIFAR-10 | FID: 17.21 -> 14.58 | Improved |
| add IEC to DeepCache N=10 | DeepCache · LSUN-Bedrooms | FID: 26.09 -> 16.91 | Improved |
| add IEC to CacheQuant W8A8 N=15 | CacheQuant · LSUN-Churches | FID: 9.47 -> 6.90 | Improved |
| add IEC to CacheQuant W8A8 N=10 | CacheQuant · ImageNet | IS: 184.38 -> 196.20 | Improved |
| add IEC to CacheQuant W8A8 N=10 | CacheQuant · MS-COCO | CLIP Score: 26.41 -> 26.45 | No clear effect |
| add IEC to CacheQuant W4A8 N=5 | CacheQuant · MS-COCO | FID: 23.85 -> 23.80 | No clear effect |
| apply IEC to all timesteps instead of none (ablation) | W8A8 quantized DDIM · CIFAR-10 | FID: 4.32 -> 3.76 | Improved |
| apply IEC only to first and last 1/20 of timesteps (ablation) | W8A8 quantized DDIM · CIFAR-10 | FID: 4.32 -> 3.97 | Improved |
| apply IEC to all timesteps instead of none (ablation) | DeepCache N=10 · CIFAR-10 | FID: 9.74 -> 7.77 | Improved |
| apply IEC only to first and last 1/10 timesteps (ablation) | DeepCache N=10 · CIFAR-10 | FID: 9.74 -> 9.58 | Improved |
Closest papers by meaning
- Optimizing Quantized Diffusion Models via Distillation with Cross-Timestep Error Correction (AAAI 2025 · similarity 0.90)
- Diffusion Model is Effectively Its Own Teacher (CVPR 2025 · similarity 0.90)
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.89)
- On Error Propagation of Diffusion Models (ICLR 2024 · similarity 0.88)
- Simple and Fast Distillation of Diffusion Models (NeurIPS 2024 · similarity 0.88)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 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.
