Research map / Architectures and efficiency / Quantization
I&S-ViT: An Inclusive & Stable Method for Post-Training ViTs Quantization
Yunshan Zhong, Jiawei Hu, Mingbao Lin, Mengzhao Chen, 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 shift-uniform-log2 quantizer (SULQ) over log2 quantizer (ablation) | DeiT-S PTQ · ImageNet W3/A3 (DeiT-S) | top-1 accuracy (%): 3.36 -> 20.70 | Improved |
| add smooth optimization strategy (SOS) (ablation) | DeiT-S PTQ · ImageNet W3/A3 (DeiT-S) | top-1 accuracy (%): 3.36 -> 45.19 | Improved |
| add SULQ on top of SOS (ablation) | DeiT-S PTQ · ImageNet W3/A3 (DeiT-S) | top-1 accuracy (%): 45.19 -> 55.78 | Improved |
| add SOS on top of SULQ (ablation) | DeiT-S PTQ · ImageNet W3/A3 (DeiT-S) | top-1 accuracy (%): 20.70 -> 55.78 | Improved |
| use SULQ instead of log2 quantizer for post-Softmax (ablation) | I&S-ViT · ImageNet W3/A3 (DeiT-S) | top-1 accuracy (%): 52.60 -> 55.78 | Improved |
| use SULQ instead of uniform quantizer for post-Softmax (ablation) | I&S-ViT · ImageNet W3/A3 (DeiT-S) | top-1 accuracy (%): 44.79 -> 55.78 | Improved |
| increase calibration images from 32 to 1024 (ablation) | I&S-ViT · ImageNet W3/A3 (DeiT-S) | top-1 accuracy (%): 39.70 -> 55.78 | Improved |
| increase calibration images from 512 to 1024 (ablation) | I&S-ViT · ImageNet W3/A3 (DeiT-S) | top-1 accuracy (%): 54.13 -> 55.78 | Improved |
| increase calibration images from 1024 to 2048 (ablation) | I&S-ViT · ImageNet W3/A3 (DeiT-S) | top-1 accuracy (%): 55.78 -> 55.89 | No clear effect |
| use I&S-ViT instead of RepQ-ViT (3-bit ViT-B) | ViT-B PTQ · ImageNet W3/A3 (ViT-B) | top-1 accuracy (%): 0.14 -> 63.77 | Improved |
| use I&S-ViT instead of RepQ-ViT (4-bit ViT-B) | ViT-B PTQ · ImageNet W4/A4 (ViT-B) | top-1 accuracy (%): 68.48 -> 80.07 | Improved |
| use I&S-ViT instead of RepQ-ViT (4-bit Swin-S detection) | Mask R-CNN Swin-S · COCO W4/A4 | AP box: 44.2 -> 43.4 | Hurt |
Later papers that compared with this method
1 later accepted paper reports I&S-ViT as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix Approximation (CVPR 2025 · better on 2 of 2 shared results)
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.
- RepQ-ViT — RepQ-ViT: Scale Reparameterization for Post-Training Quantization of Vision Transformers (ICCV 2023) · better on 7 of 7 shared results
- PD-Quant — PD-Quant: Post-Training Quantization Based on Prediction Difference Metric (CVPR 2023) · better on 1 of 1 shared result
- QDrop — QDrop: Randomly Dropping Quantization for Extremely Low-bit Post-Training Quantization (ICLR 2022) · 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
- AIQViT: Architecture-Informed Post-Training Quantization for Vision Transformers (AAAI 2025 · similarity 0.94)
- GradQ-ViT: Robust and Efficient Gradient Quantization for Vision Transformers (AAAI 2025 · similarity 0.92)
- APHQ-ViT: Post-Training Quantization with Average Perturbation Hessian Based Reconstruction for Vision Transformers (CVPR 2025 · similarity 0.91)
- GPLQ: A General, Practical, and Lightning QAT Method for Vision Transformers (NeurIPS 2025 · similarity 0.91)
- Bi-ViT: Pushing the Limit of Vision Transformer Quantization (AAAI 2024 · similarity 0.91)
- RPFQ-ViT: Rotated Phase-Frame Quantization for Extremely Low-Bit Weights in Vision Transformers (NeurIPS 2026 · similarity 0.91)
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.
