AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

BiViT: Extremely Compressed Binary Vision Transformers

Yefei He, Zhenyu Lou, Luoming Zhang, Jing Liu, Weijia Wu, Hong Zhou, Bohan Zhuang

ICCV 2023 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 42 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
replace BiBERT/BiT binarization with Softmax-aware Binarization, CLB and PWSbinary NesT-T (all binarized) · TinyImageNettop-1 accuracy (%): 32.39 -> 52.21Improved
keep MLP activations full precisionbinary NesT-T · TinyImageNettop-1 accuracy (%): 52.21 -> 69.83Improved
add Softmax-aware Binarization to attention binarization (ablation)BiBERT on NesT-T · TinyImageNettop-1 accuracy (%): 68.51 -> 70.73Improved
use softmax in the backward gradient approximation (ablation)SAB on NesT-T · TinyImageNettop-1 accuracy (%): 68.71 -> 70.73Improved
use Cross-layer Binarization instead of one-step training (ablation)binary NesT-T · TinyImageNettop-1 accuracy (%): 58.20 -> 69.83Improved
use CLB instead of traditional two-step training (ablation)binary NesT-T · TinyImageNettop-1 accuracy (%): 65.64 -> 69.83Improved
add Parameterized Weight Scales to attention (ablation)BiBERT on NesT-T · TinyImageNettop-1 accuracy (%): 68.51 -> 72.75Improved
add Parameterized Weight Scales to MLP (ablation)BiBERT on NesT-T · TinyImageNettop-1 accuracy (%): 70.02 -> 71.35Improved
lower SAB threshold coefficient beta below 0.2 (ablation)SAB on NesT-T · TinyImageNettop-1 accuracy (%): 70.73 -> 69.18Hurt
vary beta between 0.25 and 0.45 (ablation)SAB on NesT-T · TinyImageNettop-1 accuracy (%): 70.73 / 70.81 / 70.68No clear effect
use BiT learned fixed thresholds instead of BiBERT Boolbinary Swin-T / NesT-T · TinyImageNettop-1 accuracy (%): Swin-T 41.89 -> 40.52; NesT-T 32.39 -> 34.72Mixed
use local-window attention (Swin/NesT) instead of full attention (DeiT)binary ViTs · ImageNet-1k valtop-1 accuracy (%): DeiT-T 37.9 vs Swin-T 70.8Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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