Research map / Architectures and efficiency / Quantization
Post-Training Quantization for Vision Transformer
Read the paperCheck 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 ranking-aware loss to similarity-aware quantization (ablation) | ViT-B 8-bit · ImageNet | top-1 accuracy: 75.42 -> 75.94 | Helped |
| add bias correction to similarity-aware quantization (ablation) | ViT-B 8-bit · ImageNet | top-1 accuracy: 75.42 -> 75.81 | Helped |
| add ranking and bias correction together (ablation) | ViT-B 8-bit · ImageNet | top-1 accuracy: 75.42 -> 76.49 | Helped |
| add nuclear-norm mixed-precision to similarity-aware quantization (ablation) | ViT-B 8-bit · ImageNet | top-1 accuracy: 75.42 -> 76.26 | Helped |
| add ranking, bias correction and mixed-precision (full method) (ablation) | ViT-B 8-bit · ImageNet | top-1 accuracy: 76.49 -> 76.98 | Helped |
| use proposed method instead of percentile at 6-bit | ViT-B · ImageNet | top-1 accuracy: 71.58 -> 75.26 | Helped |
| use proposed method instead of percentile at 8-bit | ViT-L · CIFAR-10 | top-1 accuracy: 94.19 -> 97.90 | Helped |
| use proposed 6-bit model versus percentile 8-bit model | ViT-B · ImageNet | top-1 accuracy: 74.10 -> 75.26 | Helped |
| use proposed method instead of Bit-Split at 6-bit | DeiT-S · ImageNet | top-1 accuracy: 74.04 -> 74.58 | Helped |
| use proposed mixed-precision instead of plain 8-bit | DeiT-B · ImageNet | top-1 accuracy: 80.48 -> 81.29 | Helped |
| use proposed method instead of Bit-Split at 6-bit | DETR · COCO2017 | mAP: 38.9 -> 40.1 | Helped |
| use proposed mixed-precision instead of plain 8-bit | DETR · COCO2017 | mAP: 41.2 -> 41.7 | Helped |
Closest papers by meaning
- AIQViT: Architecture-Informed Post-Training Quantization for Vision Transformers (AAAI 2025 · similarity 0.91)
- Bi-ViT: Pushing the Limit of Vision Transformer Quantization (AAAI 2024 · similarity 0.89)
- MimiQ: Low-Bit Data-Free Quantization of Vision Transformers with Encouraging Inter-Head Attention Similarity (AAAI 2025 · similarity 0.89)
- APHQ-ViT: Post-Training Quantization with Average Perturbation Hessian Based Reconstruction for Vision Transformers (CVPR 2025 · similarity 0.89)
- FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix Approximation (CVPR 2025 · similarity 0.89)
- GradQ-ViT: Robust and Efficient Gradient Quantization for Vision Transformers (AAAI 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.