Research map / Architectures and efficiency / Quantization
Towards Accurate Post-Training Quantization of Vision Transformers via Error Reduction
Yunshan Zhong, Huang You, Jiawei Hu, Yuxin Zhang, Rongrong Ji
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 activation quantization error reduction (Aqer) (ablation) | ERQ on calibration baseline · DeiT-S W4A4 ImageNet | top-1 accuracy: 40.38 -> 70.96 | Improved |
| add weight quantization error reduction (Wqer) (ablation) | ERQ on calibration baseline · DeiT-S W4A4 ImageNet | top-1 accuracy: 40.38 -> 50.92 | Improved |
| add Wqer on top of Aqer (ablation) | ERQ · DeiT-S W4A4 ImageNet | top-1 accuracy: 70.96 -> 74.35 | Improved |
| add Reparameterization Initialization to Aqer (ablation) | Aqer · DeiT-S W4A4 ImageNet | top-1 accuracy: 50.92 -> 72.95 | Improved |
| add Ridge Regression to Aqer (ablation) | Aqer · DeiT-S W4A4 ImageNet | top-1 accuracy: 50.92 -> 59.59 | Improved |
| add Dual Uniform Quantization to Wqer (ablation) | Wqer · DeiT-S W4A4 ImageNet | top-1 accuracy: 70.96 -> 73.03 | Improved |
| add Rounding Refinement to Wqer (ablation) | Wqer · DeiT-S W4A4 ImageNet | top-1 accuracy: 70.96 -> 71.80 | Improved |
| add Ridge Regression to Wqer (ablation) | Wqer · DeiT-S W4A4 ImageNet | top-1 accuracy: 70.96 -> 72.12 | Improved |
| increase Rounding Refinement k from 1 to 3 (ablation) | Rounding Refinement · DeiT-S W4A4 ImageNet | top-1 accuracy: 74.35 -> 73.79 | Hurt |
| use 5% of D_out as outlier channels instead of none (ablation) | Dual Uniform Quantization · DeiT-S W4A4 ImageNet | top-1 accuracy: 72.76 -> 74.35 | Improved |
| increase outlier channels from 5% to 20% of D_out (ablation) | Dual Uniform Quantization · DeiT-S W4A4 ImageNet | top-1 accuracy: 74.35 -> 73.26 | Hurt |
| raise maximum iterations T from 0 to 20 (ablation) | Rounding Refinement · DeiT-S W4A4 ImageNet | top-1 accuracy: 74.06 -> 74.35 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- QDrop — QDrop: Randomly Dropping Quantization for Extremely Low-bit Post-Training Quantization (ICLR 2022) · better on 2 of 2 shared results
- DuQuant — DuQuant: Distributing Outliers via Dual Transformation Makes Stronger Quantized LLMs (NeurIPS 2024) · better on 1 of 1 shared result
- RepQ-ViT — RepQ-ViT: Scale Reparameterization for Post-Training Quantization of Vision Transformers (ICCV 2023) · better on 1 of 1 shared result
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
- ERQ: Error Reduction for Post-Training Quantization of Vision Transformers (ICML 2024 · similarity 0.99)
- AIQViT: Architecture-Informed Post-Training Quantization for Vision Transformers (AAAI 2025 · similarity 0.93)
- APHQ-ViT: Post-Training Quantization with Average Perturbation Hessian Based Reconstruction for Vision Transformers (CVPR 2025 · similarity 0.92)
- GPLQ: A General, Practical, and Lightning QAT Method for Vision Transformers (NeurIPS 2025 · similarity 0.91)
- FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix Approximation (CVPR 2025 · similarity 0.90)
- Quant Experts: Token-aware Adaptive Error Reconstruction with Mixture of Experts for Large Vision-Language Models Quantization (CVPR 2026 · 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.
