Research map / Image generation / Fast diffusion sampling
Temporal Feature Matters: A Framework for Diffusion Model Quantization
Yushi Huang, Ruihao Gong, Xianglong Liu, Jing Liu, Yuhang Li, Jiwen Lu, Dacheng Tao
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 temporal information-aware reconstruction (TIAR) (ablation) | BRECQ baseline · LSUN-Bedrooms 256x256 W4A8 (LDM-4) | FID: 9.36 -> 4.84 | Improved |
| add finite set calibration (FSC) (ablation) | BRECQ baseline · LSUN-Bedrooms 256x256 W4A8 (LDM-4) | FID: 9.36 -> 6.07 | Improved |
| add FSC on top of TIAR (ablation) | TIAR · LSUN-Bedrooms 256x256 W4A8 (LDM-4) | FID: 4.84 -> 3.68 | Improved |
| use cache-based maintenance instead of baseline (ablation) | BRECQ baseline · LSUN-Bedrooms 256x256 W4A8 (LDM-4) | FID: 9.36 -> 3.91 | Improved |
| add disturbance-aware selection combining TM and CM (ablation) | TM/CM · LSUN-Bedrooms 256x256 W4A8 (LDM-4) | FID: 3.68 -> 3.61 | Improved |
| use LSQ instead of Min-Max for cache calibration (ablation) | Cache-based Maintenance · LSUN-Bedrooms 256x256 W4A8 (LDM-4) | FID: 5.41 -> 3.91 | Improved |
| use MSE/LSQ instead of Min-Max for TM calibration (ablation) | TIB-based Maintenance · LSUN-Bedrooms 256x256 W4A8 (LDM-4) | FID: 3.68 -> 3.62 | No clear effect |
| replace MSE loss with KL or CE in selection (ablation) | Disturbance-aware Selection · LSUN-Bedrooms 256x256 W4A8 (LDM-4) | FID: 3.61 -> 3.63, 3.64 | No clear effect |
| apply DS vs. Q-Diffusion/PTQD baselines | Quantized LDM · CelebA-HQ 256x256 W4A8 | FID: 15.55 -> 8.73 | Improved |
| apply DS instead of Q-Diffusion | Quantized SD-XL-base · MS-COCO W4A8 | FID: 31.18 -> 25.57 | Improved |
| apply DS instead of Q-Diffusion | Quantized SD-XL-turbo · MS-COCO W4A8 | FID: 23.71 -> 18.21 | Improved |
| apply DS instead of Q-Diffusion | Quantized Stable Diffusion-v1-4 · MS-COCO W4A8 | FID: 14.49 -> 13.29 | 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.
- PTQD — PTQD: Accurate Post-Training Quantization for Diffusion Models (NeurIPS 2023) · better on 2 of 2 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
- TFMQ-DM: Temporal Feature Maintenance Quantization for Diffusion Models (CVPR 2024 · similarity 0.94)
- TCAQ-DM: Timestep-Channel Adaptive Quantization for Diffusion Models (AAAI 2025 · similarity 0.91)
- Towards Accurate Post-training Quantization for Diffusion Models (CVPR 2024 · similarity 0.90)
- EfficientDM: Efficient Quantization-Aware Fine-Tuning of Low-Bit Diffusion Models (ICLR 2024 · similarity 0.90)
- Optimizing Quantized Diffusion Models via Distillation with Cross-Timestep Error Correction (AAAI 2025 · similarity 0.89)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · 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.
