AI Paper Atlas

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

ECCV 2022 · Architectures and efficiency › Attention and transformers · approach: Hybrid and interpretable ViT designs · 69 citations (OpenAlex)

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 changedWhereMeasuredEffect
add L_PSE patch similarity entropy loss to L_OH and L_TV (ablation)PSAQ-ViT DeiT-S · ImageNet W8/A8top-1 accuracy (%): 65.66 -> 76.92Improved
add L_PSE patch similarity entropy loss to L_OH and L_TV (ablation)PSAQ-ViT DeiT-B · ImageNet W8/A8top-1 accuracy (%): 67.95 -> 79.10Improved
optimize noise with L_PSE alone instead of Gaussian noise (ablation)PSAQ-ViT DeiT-S · ImageNet W8/A8top-1 accuracy (%): 18.16 -> 74.07Improved
add L_OH to L_PSE (ablation)PSAQ-ViT DeiT-S · ImageNet W8/A8top-1 accuracy (%): 74.07 -> 75.39Improved
add L_TV to L_PSE+L_OH (ablation)PSAQ-ViT DeiT-S · ImageNet W8/A8top-1 accuracy (%): 75.39 -> 76.92Improved
add L_TV and L_OH to L_PSE (ablation)PSAQ-ViT DeiT-B · ImageNet W8/A8top-1 accuracy (%): 78.07 -> 79.10Improved
calibrate with PSAQ-ViT generated images instead of Gaussian noiseSwin-T W4/A8 · ImageNettop-1 accuracy (%): 0.52 -> 71.79Improved
calibrate with PSAQ-ViT images instead of real images (Standard)DeiT-S W4/A8 · ImageNettop-1 accuracy (%): 72.10 -> 73.23Improved
calibrate with PSAQ-ViT images instead of real images (Standard)ViT-S W8/A8 · ImageNettop-1 accuracy (%): 30.28 -> 31.45Improved
combine PSAQ-ViT with OMSE instead of vanilla calibrationDeiT-B W8/A8 · ImageNettop-1 accuracy (%): 79.10 -> 80.26Improved
combine PSAQ-ViT with Percentile instead of vanilla calibrationDeiT-S W8/A8 · ImageNettop-1 accuracy (%): 76.92 -> 77.31Improved
combine PSAQ-ViT with OMSE instead of vanilla calibrationDeiT-S W8/A8 · ImageNettop-1 accuracy (%): 76.92 -> 76.94No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy