Research map / Architectures and efficiency / Quantization
Diversifying Sample Generation for Accurate Data-Free Quantization
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 DSG-generated diversified calibration data | ZeroQ data generation for ResNet-18 quantization · ImageNet, ResNet-18 W4A4 | Top-1 accuracy: 26.04 -> 34.53 | Helped |
| Add DSG-generated diversified calibration data | ZeroQ data generation for SqueezeNext quantization · ImageNet, SqueezeNext W6A6 | Top-1 accuracy: 39.83 -> 60.50 | Helped |
| Add DSG-generated diversified calibration data | ZeroQ data generation for InceptionV3 quantization · ImageNet, InceptionV3 W4A4 | Top-1 accuracy: 12.00 -> 34.89 | Helped |
| Add Slack Distribution Alignment (SDA) (ablation) | DSG, ResNet-18 · ImageNet W4A4 | Top-1 accuracy: 26.04 -> 33.39 | Helped |
| Add Layerwise Sample Enhancement (LSE) (ablation) | DSG, ResNet-18 · ImageNet W4A4 | Top-1 accuracy: 26.04 -> 27.12 | Helped |
| Combine SDA and LSE (ablation) | DSG, ResNet-18 · ImageNet W4A4 | Top-1 accuracy: 26.04 -> 34.53 | Helped |
| Increase SDA relaxation percentile from 0.9 to 1 (ablation) | DSG, ResNet-18 · ImageNet W4A4 | Top-1 accuracy | Hurt |
| Use DSG rather than ZeroQ-generated data | AdaRound, ResNet-18 · ImageNet W3A32 | Top-1 accuracy: 49.86 -> 56.09 | Helped |
| Add labels to DSG-generated data (ablation) | AdaRound, ResNet-18 · ImageNet W3A32 | Top-1 accuracy: 56.09 -> 58.27 | Helped |
| Add image prior to labeled DSG-generated data (ablation) | AdaRound, ResNet-18 · ImageNet W3A32 | Top-1 accuracy: 58.27 -> 61.32 | Helped |
Closest papers by meaning
- Enhancing Diversity for Data-free Quantization (CVPR 2025 · similarity 0.92)
- Adaptive Dataset Quantization (AAAI 2025 · similarity 0.87)
- Quality-Diversity Generative Sampling for Learning with Synthetic Data (AAAI 2024 · similarity 0.86)
- QuantDemoire: Quantization with Outlier Aware for Image Demoiréing (NeurIPS 2026 · similarity 0.86)
- Sharpness-Aware Data Generation for Zero-shot Quantization (ICML 2024 · similarity 0.86)
- Robustness-Guided Image Synthesis for Data-Free Quantization (AAAI 2024 · similarity 0.85)
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.