Research map / Architectures and efficiency / Quantization
Temporal Dynamic Quantization for Diffusion Models
Junhyuk So, Jungwon Lee, Daehyun Ahn, Hyungjun Kim, Eunhyeok Park
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 TDQ module to LSQ | LSQ · CIFAR-10 DDIM W4A4 QAT | FID: 7.3 -> 4.48 | Improved |
| add TDQ module to LSQ | LSQ · CIFAR-10 DDIM W3A3 QAT | FID: 7.63 -> 6.48 | Improved |
| add TDQ module to NIPQ | NIPQ · CIFAR-10 DDIM W4A4 | FID: 30.73 -> 24.27 | Improved |
| add TDQ module to DuQ | DuQ · CIFAR-10 DDIM | FID: 7.88 -> 4.46 | Improved |
| add TDQ module to QIL | QIL · CIFAR-10 DDIM | FID: 9.85 -> 7.51 | Improved |
| add TDQ module to PACT | PACT · CIFAR-10 DDIM | FID: 51.92 -> 74.56 | Hurt |
| replace Min-Max PTQ with TDQ at W8A6 | Min-Max · CIFAR-10 DDIM PTQ | FID: 332.15 -> 5.71 | Improved |
| replace dynamic quantization with time-based TDQ (ablation) | dynamic quantization · CIFAR-10 DDIM W4A4 QAT | FID: 6.21 -> 4.48 | Improved |
| add layer statistics generator to dynamic quantization (ablation) | dynamic quantization · CIFAR-10 DDIM W4A4 QAT | FID: 6.21 -> 5.49 | Improved |
| add time information to layer-statistics generator (ablation) | dynamic quantization · CIFAR-10 DDIM W4A4 QAT | FID: 5.49 -> 4.96 | Improved |
| use frequency encoding instead of normalizing time step (ablation) | TDQ · CIFAR-10 DDIM W4A4 QAT | FID: 5.03 -> 4.48 | Improved |
| use Zero initialization instead of He (ablation) | TDQ weights · CIFAR-10 W4A4 | FID: 4.48 -> 4.98 | Hurt |
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.
- NIPQ — NIPQ: Noise Proxy-Based Integrated Pseudo-Quantization (CVPR 2023) · better on 7 of 8 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 0.91)
- Q-DiT: Accurate Post-Training Quantization for Diffusion Transformers (CVPR 2025 · similarity 0.90)
- TCAQ-DM: Timestep-Channel Adaptive Quantization for Diffusion Models (AAAI 2025 · similarity 0.90)
- Optimizing Quantized Diffusion Models via Distillation with Cross-Timestep Error Correction (AAAI 2025 · similarity 0.90)
- EfficientDM: Efficient Quantization-Aware Fine-Tuning of Low-Bit Diffusion Models (ICLR 2024 · similarity 0.89)
- ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generation (ICLR 2025 · similarity 0.89)
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.
