AI Paper Atlas

Research map / Architectures and efficiency / Quantization

LS-ViT: Least-Squares Hessian Based Block Reconstruction for Low-Bit Post-Training Quantization of Vision Transformers

Hyunha Hwang, Xuan Truong Nguyen, Hyuk-Jae Lee

CVPR 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of ViTs

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
replace averaging-based diagonal Hessian (FIMA-Q-D) with least-squares diagonal Hessian (ablation)Hessian-based block reconstruction · ImageNet (W3/A3), ViT-STop-1 accuracy (%): 60.02 -> 63.25Improved
replace averaging-based diagonal with least-squares diagonal on Swin-S (ablation)Hessian-based block reconstruction · ImageNet (W3/A3), Swin-STop-1 accuracy (%): 75.08 -> 77.29Improved
drop the division by dz_i and use g_i alone instead of g_i/dz_i (ablation)Sample-wise diagonal Hessian · ImageNet (W3/A3), ViT-STop-1 accuracy (%): 23.59 -> 58.16Improved
use averaging-based FIMA-Q-D instead of sample-wise g_i (ablation)Diagonal Hessian approximation · ImageNet (W3/A3), Swin modelsTop-1 accuracy (%): +0.68%p average over g_i; Swin-S 76.02 -> 75.08Mixed
reduce FIMA-Q Hessian estimation from 15 to 1 full-model computation (ablation)FIMA-Q · ImageNet (W3/A3), Swin-STop-1 accuracy (%): 77.26 -> 75.85Hurt
use LS-ViT with a single full-model computation instead of FIMA-Q with 15Hessian estimation · ImageNet (W3/A3), Swin-STop-1 accuracy (%): 77.26 -> 77.39Improved
replace FIMA-Q with LS-ViT at W2/A3Swin-S · ImageNet (W2/A3), Swin-STop-1 accuracy (%): 59.57 -> 65.77Improved
replace FIMA-Q with LS-ViT at W3/A3seven ViT/DeiT/Swin models · ImageNet (W3/A3)Top-1 accuracy (%): average +0.17%pNo clear effect
replace FIMA-Q with LS-ViT at W2/A4seven ViT/DeiT/Swin models · ImageNet (W2/A4)Top-1 accuracy (%): average +1.46%pImproved
replace FIMA-Q with LS-ViT on COCO detectionMask R-CNN Swin-T · COCO (W4/A4)APb: 38.7 -> 38.9Improved
use LS-ViT instead of FIMA-Q for quantization trainingQuantization time · Swin-B training timetime (min): 480 -> 180Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.

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

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