AI Paper Atlas

Research map / Architectures and efficiency / Quantization

AdaLog: Post-Training Quantization for Vision Transformers with Adaptive Logarithm Quantizer

Zhuguanyu Wu, Jiaxin Chen, Hanwen Zhong, Di Huang, Yunhong Wang

ECCV 2024 · Architectures and efficiency › Quantization · approach: Post-training quantization of ViTs · 24 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 AdaLog quantizer on post-Softmax and post-GELU activations (ablation)ViT-S PTQ · ImageNet W4/A4top-1 accuracy: 62.20 -> 72.01Improved
add FPCS hyperparameter search (ablation)ViT-B PTQ · ImageNet W3/A3top-1 accuracy: 9.68 -> 15.50Improved
combine AdaLog with FPCS (ablation)ViT-B PTQ · ImageNet W3/A3top-1 accuracy: 9.68 -> 37.91Improved
add FPCS on top of AdaLog for DeiT-S (ablation)DeiT-S PTQ · ImageNet W3/A3top-1 accuracy: 22.81 -> 24.47No clear effect
replace RepQ-ViT quantizers with AdaLog at 4-bitRepQ-ViT · ImageNet W4/A4 averagetop-1 accuracy: average +5.13%Improved
use AdaLog at 3-bit instead of RepQ-ViTRepQ-ViT · ImageNet W3/A3 ViT-Stop-1 accuracy: 0.10 -> 13.88Improved
replace log-sqrt2 quantizer with AdaLog (lookup tables, integer-only)DeiT-T W4/A4 · inference costFixOPs: 0.613B -> 0.539BImproved
train activation scaling with LSQ in BRECQ with AdaLog (ablation)BRECQ ViT-S · ImageNet W4/A4top-1 accuracy: 76.48 -> 77.25Improved
train activation scaling with LSQ in BRECQ without AdaLog (ablation)BRECQ ViT-S · ImageNet W4/A4top-1 accuracy: 69.21 -> 1.93Hurt
replace alternating search with FPCS (ablation)AdaLog calibration · DeiT-T W3A3top-1 accuracy: 28.41 -> 31.56Improved
replace brute force search with FPCS (ablation)AdaLog calibration · DeiT-T W3A3GPU minutes: 183 -> 4.1Improved

Later papers that compared with this method

2 later accepted papers report AdaLog as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.

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