Research map / Architectures and efficiency / Quantization
Diverse Sample Generation: Pushing the Limit of Generative Data-Free Quantization
Haotong Qin, Yifu Ding, Xiangguo Zhang, Jiakai Wang, Xianglong Liu, Jiwen Lu
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 Slack Distribution Alignment (SDA) (ablation) | ZeroQ vanilla · ImageNet ResNet18 W4A4 PTQ | top-1 accuracy: 26.04 -> 33.39 | Improved |
| add Slack Distribution Alignment (SDA) (ablation) | GDFQ vanilla · ImageNet ResNet18 W4A4 QAT | top-1 accuracy: 60.52 -> 61.08 | Improved |
| set slack epsilon to 1 instead of 0.9 (ablation) | SDA in PTQ · ImageNet ResNet18 W4A4 PTQ | top-1 accuracy: drops 3.61% | Hurt |
| set slack epsilon to 1 instead of 0.9 (ablation) | SDA in QAT · ImageNet ResNet18 W4A4 QAT | top-1 accuracy: drops 0.02% | Hurt |
| add Layerwise Sample Enhancement (LSE) (ablation) | ZeroQ vanilla · ImageNet ResNet18 W4A4 PTQ | top-1 accuracy: 26.04 -> 27.12 | Improved |
| add Layerwise Sample Enhancement (LSE) (ablation) | GDFQ vanilla · ImageNet ResNet18 W4A4 QAT | top-1 accuracy: 60.52 -> 60.71 | Improved |
| add Sample Correlation Inhibition (SCI) (ablation) | ZeroQ vanilla · ImageNet ResNet18 W4A4 PTQ | top-1 accuracy: 26.04 -> 39.53 | Improved |
| add Sample Correlation Inhibition (SCI) (ablation) | GDFQ vanilla · ImageNet ResNet18 W4A4 QAT | top-1 accuracy: 60.52 -> 60.92 | Improved |
| combine SDA, LSE and SCI (full DSG) (ablation) | GDFQ vanilla · ImageNet ResNet18 W4A4 QAT | top-1 accuracy: 60.52 -> 62.18 | Improved |
| combine SDA, LSE and SCI (full DSG) (ablation) | ZeroQ vanilla · ImageNet ResNet18 W4A4 PTQ | top-1 accuracy: 26.04 -> 39.90 | Improved |
| calibrate with DSG synthetic data instead of real data | ResNet20 CIFAR10 PTQ · CIFAR10 W4A4 | top-1 accuracy: 87.38 -> 87.79 | Improved |
| replace ZeroQ synthetic data with DSG data and SQuant calibration | ResNet18 · ImageNet PTQ W4A4 | top-1 accuracy: 65.22 -> 66.67 | 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.
- SQuant — SQuant: On-the-Fly Data-Free Quantization via Diagonal Hessian Approximation (ICLR 2022) · better on 4 of 4 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
- Enhancing Diversity for Data-free Quantization (CVPR 2025 · similarity 0.91)
- Diverse Rare Sample Generation with Pretrained GANs (AAAI 2025 · similarity 0.88)
- Adaptive Dataset Quantization (AAAI 2025 · similarity 0.87)
- Efficient Generative Modeling with Residual Vector Quantization-Based Tokens (ICML 2025 · similarity 0.87)
- Robustness-Guided Image Synthesis for Data-Free Quantization (AAAI 2024 · similarity 0.86)
- Sharpness-Aware Data Generation for Zero-shot Quantization (ICML 2024 · similarity 0.86)
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.
