Research map / Image generation / Fast diffusion sampling
Gradient-Aligned Calibration for Post-Training Quantization of Diffusion Models
Dung Anh Hoang, Cuong Pham, Trung Le, Jianfei Cai, Thanh-Toan Do
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 |
|---|---|---|---|
| learn gradient-aligned sample weights for calibration | TFMQ-DM · CIFAR-10 W4A32 | FID: 4.73 -> 4.28 | Improved |
| learn gradient-aligned sample weights | TFMQ-DM · LSUN-Bedrooms W4A32 | FID: 3.60 -> 3.14 | Improved |
| learn gradient-aligned sample weights | TFMQ-DM · ImageNet W4A32 | FID: 10.50 -> 10.17 | Improved |
| learn gradient-aligned sample weights | TFMQ-DM · ImageNet W4A8 | FID/sFID: sFID 7.35 -> 7.55; FID 10.29 -> 9.96 | Mixed |
| apply to extreme 2/4-bit quantization | TFMQ-DM · ImageNet W2A4 | FID: 300.03 -> 226.27 | Improved |
| apply with few sampling steps (5) | TFMQ-DM · ImageNet W4A32, 5 steps | FID: 19.10 -> 18.22 | Improved |
| lower temperature from 1 to 0.2 (ablation) | Ours · CIFAR-10 W4A32 | FID: 4.28 -> 4.85 | Hurt |
| enlarge validation set from 5% to 20% (ablation) | Ours · CIFAR-10 W4A32 | FID: 4.32 -> 4.75 | Hurt |
| more timestep groups (5 to 20) (ablation) | Ours · CIFAR-10 W4A32 | FID: 4.28 -> 4.26 | No clear effect |
| use a single group (uniform task) instead of 5 (ablation) | Ours · CIFAR-10 W4A32 | FID: 4.63 -> 4.28 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TFMQ-DM — TFMQ-DM: Temporal Feature Maintenance Quantization for Diffusion Models (CVPR 2024) · better on 11 of 12 shared results
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
- Towards Accurate Post-training Quantization for Diffusion Models (CVPR 2024 · similarity 0.93)
- EfficientDM: Efficient Quantization-Aware Fine-Tuning of Low-Bit Diffusion Models (ICLR 2024 · similarity 0.91)
- TCAQ-DM: Timestep-Channel Adaptive Quantization for Diffusion Models (AAAI 2025 · similarity 0.91)
- Q-DiT: Accurate Post-Training Quantization for Diffusion Transformers (CVPR 2025 · similarity 0.91)
- TFMQ-DM: Temporal Feature Maintenance Quantization for Diffusion Models (CVPR 2024 · similarity 0.91)
- AccuQuant: Simulating Multiple Denoising Steps for Quantizing Diffusion Models (NeurIPS 2025 · similarity 0.91)
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.
