Research map / Image generation / Fast diffusion sampling
PTQD: Accurate Post-Training Quantization for Diffusion Models
Yefei He, Luping Liu, Jing Liu, Weijia Wu, Hong Zhou, Bohan Zhuang
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 |
|---|---|---|---|
| Apply PTQD corrections to 4/8-bit quantization | LDM-4 · ImageNet 256×256, 250 steps, η=1.0, scale=1.5 | FID: 5.37 -> 5.11 | Improved |
| Apply PTQD corrections to mixed precision | LDM-8 · LSUN-Churches 256×256, 200 steps, η=0.0 | FID: 218.59 -> 17.99 | Improved |
| Add Correlated Noise Correction (ablation) | LDM-4 with Q-Diffusion mixed precision · ImageNet 256×256, 250 steps | FID: 9.97 -> 9.49 | Improved |
| Add Variance Schedule Calibration after Correlated Noise Correction (ablation) | LDM-4 with Q-Diffusion mixed precision · ImageNet 256×256, 250 steps | FID: 9.49 -> 9.29 | Improved |
| Add Bias Correction after Correlated Noise Correction and Variance Schedule Calibration (ablation) | LDM-4 with Q-Diffusion mixed precision · ImageNet 256×256, 250 steps | FID: 9.29 -> 6.44 | Improved |
| Add Correlated Noise Correction (ablation) | LDM-4 with Q-Diffusion 4/8-bit quantization · LSUN-Bedrooms 256×256, 200 steps | FID: 6.72 -> 6.31 | Improved |
| Add Variance Schedule Calibration after Correlated Noise Correction (ablation) | LDM-4 with Q-Diffusion 4/8-bit quantization · LSUN-Bedrooms 256×256, 200 steps | FID: 6.31 -> 6.10 | Improved |
| Add Bias Correction after Correlated Noise Correction and Variance Schedule Calibration (ablation) | LDM-4 with Q-Diffusion 4/8-bit quantization · LSUN-Bedrooms 256×256, 200 steps | FID: 6.10 -> 5.94 | Improved |
| Use PTQD instead of Q-Diffusion for 4/8-bit quantization | LDM-4 · LSUN-Bedrooms 256×256, 200 steps | FID: 6.72 -> 5.94 | Improved |
| Use PTQD instead of Q-Diffusion for mixed precision | LDM-4 · ImageNet 256×256, 20 steps, η=0.0 | FID: 116.61 -> 7.75 | Improved |
Later papers that compared with this method
7 later accepted papers report PTQD as a baseline in their results tables; 7 of them report a better number on most of the results they share with it.
- Sampling-Aware Quantization for Diffusion Models (CVPR 2026 · better on 5 of 5 shared results)
- D^2-DPM: Dual Denoising for Quantized Diffusion Probabilistic Models (AAAI 2025 · better on 10 of 10 shared results)
- Temporal Feature Matters: A Framework for Diffusion Model Quantization (TPAMI 2025 · better on 2 of 2 shared results)
- AccuQuant: Simulating Multiple Denoising Steps for Quantizing Diffusion Models (NeurIPS 2025 · better on 1 of 1 shared result)
- StepbaQ: Stepping backward as Correction for Quantized Diffusion Models (NeurIPS 2024 · better on 11 of 12 shared results)
- TFMQ-DM: Temporal Feature Maintenance Quantization for Diffusion Models (CVPR 2024 · better on 6 of 6 shared results)
- MixDQ: Memory-Efficient Few-Step Text-to-Image Diffusion Models with Metric-Decoupled Mixed Precision Quantization (ECCV 2024 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- EfficientDM: Efficient Quantization-Aware Fine-Tuning of Low-Bit Diffusion Models (ICLR 2024 · similarity 0.93)
- AccuQuant: Simulating Multiple Denoising Steps for Quantizing Diffusion Models (NeurIPS 2025 · similarity 0.93)
- Towards Accurate Post-training Quantization for Diffusion Models (CVPR 2024 · similarity 0.93)
- D^2-DPM: Dual Denoising for Quantized Diffusion Probabilistic Models (AAAI 2025 · similarity 0.93)
- Q-DiT: Accurate Post-Training Quantization for Diffusion Transformers (CVPR 2025 · similarity 0.92)
- TFMQ-DM: Temporal Feature Maintenance Quantization for Diffusion Models (CVPR 2024 · similarity 0.92)
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.
