AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

MB-TaylorFormer: Multi-Branch Efficient Transformer Expanded by Taylor Formula for Image Dehazing

Yuwei Qiu, Kaihao Zhang, Chenxi Wang, Wenhan Luo, Hongdong Li, Zhi Jin

ICCV 2023 · Architectures and efficiency › Attention and transformers · approach: Efficient image transformers · 295 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 multi-scale attention refinement (MSAR) module (ablation)MB-TaylorFormer-B · SOTS-IndoorPSNR: 38.74 -> 40.71Improved
use deformable depthwise separable serial convs (DSDCN-S) in multi-scale patch embedding instead of single-branch conv (ablation)MB-TaylorFormer-B · SOTS-IndoorPSNR: 38.27 -> 40.71Improved
use deformable convs instead of standard convs in serial patch embedding (ablation)MB-TaylorFormer-B · SOTS-IndoorPSNR: 39.04 -> 40.71Improved
enlarge gating tensor G to C channels (ablation)MSAR · SOTS-IndoorPSNR: 40.71 -> 40.88, params 2.677M -> 4.288MNo clear effect
use Taylor expanded self-attention instead of other linear attentions (ablation)MB-TaylorFormer-B · SOTS-IndoorPSNR: 38.57 (MDTA) -> 40.71Improved
second-order instead of first-order Taylor expansion in Swin attention (ablation)Swin · SOTS-IndoorPSNR: 36.37 -> 36.50No clear effect
Taylor approximation versus exact softmax in Swin (ablation)Swin · SOTS-IndoorPSNR: 36.59 -> 36.37Hurt
MB-TaylorFormer-B versus SGID-PFF on SOTS-IndoorMB-TaylorFormer-B · SOTS-IndoorPSNR: 38.52 -> 40.71Improved

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