Research map / Architectures and efficiency / Attention and transformers
Patch Similarity Aware Data-Free Quantization for Vision Transformers
Zhikai Li, Liping Ma, Mengjuan Chen, Junrui Xiao, Qingyi Gu
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 L_PSE patch similarity entropy loss to L_OH and L_TV (ablation) | PSAQ-ViT DeiT-S · ImageNet W8/A8 | top-1 accuracy (%): 65.66 -> 76.92 | Improved |
| add L_PSE patch similarity entropy loss to L_OH and L_TV (ablation) | PSAQ-ViT DeiT-B · ImageNet W8/A8 | top-1 accuracy (%): 67.95 -> 79.10 | Improved |
| optimize noise with L_PSE alone instead of Gaussian noise (ablation) | PSAQ-ViT DeiT-S · ImageNet W8/A8 | top-1 accuracy (%): 18.16 -> 74.07 | Improved |
| add L_OH to L_PSE (ablation) | PSAQ-ViT DeiT-S · ImageNet W8/A8 | top-1 accuracy (%): 74.07 -> 75.39 | Improved |
| add L_TV to L_PSE+L_OH (ablation) | PSAQ-ViT DeiT-S · ImageNet W8/A8 | top-1 accuracy (%): 75.39 -> 76.92 | Improved |
| add L_TV and L_OH to L_PSE (ablation) | PSAQ-ViT DeiT-B · ImageNet W8/A8 | top-1 accuracy (%): 78.07 -> 79.10 | Improved |
| calibrate with PSAQ-ViT generated images instead of Gaussian noise | Swin-T W4/A8 · ImageNet | top-1 accuracy (%): 0.52 -> 71.79 | Improved |
| calibrate with PSAQ-ViT images instead of real images (Standard) | DeiT-S W4/A8 · ImageNet | top-1 accuracy (%): 72.10 -> 73.23 | Improved |
| calibrate with PSAQ-ViT images instead of real images (Standard) | ViT-S W8/A8 · ImageNet | top-1 accuracy (%): 30.28 -> 31.45 | Improved |
| combine PSAQ-ViT with OMSE instead of vanilla calibration | DeiT-B W8/A8 · ImageNet | top-1 accuracy (%): 79.10 -> 80.26 | Improved |
| combine PSAQ-ViT with Percentile instead of vanilla calibration | DeiT-S W8/A8 · ImageNet | top-1 accuracy (%): 76.92 -> 77.31 | Improved |
| combine PSAQ-ViT with OMSE instead of vanilla calibration | DeiT-S W8/A8 · ImageNet | top-1 accuracy (%): 76.92 -> 76.94 | No clear effect |
Closest papers by meaning
- MimiQ: Low-Bit Data-Free Quantization of Vision Transformers with Encouraging Inter-Head Attention Similarity (AAAI 2025 · similarity 0.91)
- AIQViT: Architecture-Informed Post-Training Quantization for Vision Transformers (AAAI 2025 · similarity 0.88)
- Bi-ViT: Pushing the Limit of Vision Transformer Quantization (AAAI 2024 · similarity 0.88)
- GPLQ: A General, Practical, and Lightning QAT Method for Vision Transformers (NeurIPS 2025 · similarity 0.88)
- Instance-Aware Group Quantization for Vision Transformers (CVPR 2024 · similarity 0.88)
- RPFQ-ViT: Rotated Phase-Frame Quantization for Extremely Low-Bit Weights in Vision Transformers (NeurIPS 2026 · 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.
