Research map / Image generation / Fast diffusion sampling
Learning Quantized Adaptive Conditions for Diffusion Models
Yuchen Liang, Yuchuan Tian, Lei Yu, Huaao Tang, Jie Hu, Xiangzhong Fang, Hanting Chen
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 quantized adaptive conditions (ablation) | Rectified Flow · CIFAR-10 Heun NFE 5 | FID: 37.19 -> 19.68 | Improved |
| add QAC on top of RNE (ablation) | Rectified Flow · CIFAR-10 Heun NFE 5 | FID: 19.68 -> 18.20 | Improved |
| use online instead of offline sampling weight collection (ablation) | QAC · CIFAR-10 RK45 | FID: 5.96 -> 2.61 | Improved |
| use FSQ instead of vanilla VAE condition encoder (ablation) | QAC · CIFAR-10 Heun NFE 9 | FID: 12.23 -> 10.83 | Improved |
| use smaller FSQ level (L=2 instead of L=8) at fixed codebook size (ablation) | QAC · MNIST Euler NFE 4 | FID: 8.50 -> 6.36 | Improved |
| scale codebook size from 2^12 to 2^20 (ablation) | QAC · CIFAR-10 Heun NFE 5 | FID: 21.86 -> 19.68 | Improved |
| add QAC to EDM with iPNDM at few NFE | EDM · AFHQv2 64x64 NFE 3 | FID: 15.60 -> 8.32 | Improved |
| add QAC to EDM with iPNDM at 6 NFE | EDM · CIFAR-10 NFE 6 | FID: 5.32 -> 5.14 | Improved |
| use adaptive condition instead of RNE for trajectory straightness | Rectified Flow · CIFAR-10 curvature | average curvature: 0.38 (RNE) vs 0.40 | No clear effect |
Closest papers by meaning
- On the Trajectory Regularity of ODE-based Diffusion Sampling (ICML 2024 · similarity 0.89)
- Towards Accurate Post-training Quantization for Diffusion Models (CVPR 2024 · similarity 0.89)
- Optimizing Quantized Diffusion Models via Distillation with Cross-Timestep Error Correction (AAAI 2025 · similarity 0.89)
- TCAQ-DM: Timestep-Channel Adaptive Quantization for Diffusion Models (AAAI 2025 · similarity 0.89)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.89)
- Q-DiT: Accurate Post-Training Quantization for Diffusion Transformers (CVPR 2025 · 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.
