Research map / Architectures and efficiency / Quantization
Oscillation-free Quantization for Low-bit Vision Transformers
Shih-yang Liu, Zechun Liu, Kwang-Ting Cheng
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 |
|---|---|---|---|
| Use OFQ instead of the strongest listed baseline | DeiT-T · ImageNet-1K W2A2 | Top-1 accuracy: 54.45 -> 64.33 | Improved |
| Use OFQ instead of the strongest listed baseline | DeiT-S · ImageNet-1K W2A2 | Top-1 accuracy: 68.67 -> 75.72 | Improved |
| Use OFQ instead of the strongest listed baseline | Swin-T · ImageNet-1K W2A2 | Top-1 accuracy: 73.88 -> 78.52 | Improved |
| Replace LSQ with StatsQ (ablation) | DeiT-S · ImageNet-1K W2A2 | Top-1 accuracy: 68 -> 74.3 | Improved |
| Add QKR to StatsQ (ablation) | DeiT-S · ImageNet-1K W2A2 | Top-1 accuracy: 74.3 -> 75.00 | Improved |
| Add CGA to StatsQ (ablation) | DeiT-S · ImageNet-1K W2A2 | Top-1 accuracy: 74.3 -> 75.07 | Improved |
| Add both QKR and CGA to StatsQ (ablation) | DeiT-S · ImageNet-1K W2A2 | Top-1 accuracy: 68 -> 75.72 | Improved |
| Replace LSQ with StatsQ (ablation) | DeiT-S · ImageNet-1K W3A3 | Top-1 accuracy: 77.76 -> 78.56 | Improved |
| Add QKR to StatsQ (ablation) | DeiT-S · ImageNet-1K W3A3 | Top-1 accuracy: 78.56 -> 79.15 | Improved |
| Add CGA to StatsQ (ablation) | DeiT-S · ImageNet-1K W3A3 | Top-1 accuracy: 78.56 -> 79 | Improved |
| Replace LSQ with StatsQ (ablation) | DeiT-S · ImageNet-1K W4A4 | Top-1 accuracy: 79.66 -> 80.68 | Improved |
| Add QKR to StatsQ (ablation) | DeiT-S · ImageNet-1K W4A4 | Top-1 accuracy: 80.68 -> 81.00 | Improved |
Closest papers by meaning
- Overcoming Oscillations in Quantization-Aware Training (ICML 2022 · similarity 0.92)
- AIQViT: Architecture-Informed Post-Training Quantization for Vision Transformers (AAAI 2025 · similarity 0.88)
- Zero-Shot Quantization via Weight-Space Arithmetic (NeurIPS 2026 · similarity 0.88)
- Bi-ViT: Pushing the Limit of Vision Transformer Quantization (AAAI 2024 · similarity 0.88)
- ERQ: Error Reduction for Post-Training Quantization of Vision Transformers (ICML 2024 · similarity 0.88)
- GPLQ: A General, Practical, and Lightning QAT Method for Vision Transformers (NeurIPS 2025 · similarity 0.87)
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.
