AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Axial-DeepLab: Stand-Alone Axial-Attention for Panoptic Segmentation

Huiyu Wang, Yukun Zhu, Bradley Green, Hartwig Adam, Alan Yuille, Liang-Chieh Chen

ECCV 2020 · Architectures and efficiency › Attention and transformers · approach: Theory of self-attention dynamics · 850 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 batch normalization to attention layers (ablation)Conv-Stem + Attention · ImageNet valtop-1 (%): 77.4 -> 77.7Improved
add position-sensitive self-attention (ablation)Conv-Stem + Attention · ImageNet valtop-1 (%): 77.7 -> 78.1Improved
factorize into axial-attention instead of PS-Attention (conv stem) (ablation)Conv-Stem PS-Attention · ImageNet valtop-1 (%): 78.1 -> 77.5No clear effect
use Full Axial-Attention instead of Full Attention (ablation)Full Attention · ImageNet valtop-1 (%): 77.6 -> 78.1Improved
add ASPP (ablation)ResNet-50 backbone · Cityscapes valPQ (%): 58.1 -> 59.8Improved
add ASPP (ablation)PS-Attention backbone · Cityscapes valPQ (%): 59.9 -> 61.5Improved
add position-sensitivity to attention (ablation)Attention backbone · Cityscapes valPQ (%): 58.7 -> 59.9Improved
add axial-attention instead of PS-Attention, no ASPP (ablation)PS-Attention · Cityscapes valPQ (%): 59.9 -> 62.6Improved
increase axial span from 5x5 to 65x65 (ablation)Axial-ResNet-L · Cityscapes valPQ (%): 59.1 -> 64.2Improved
increase output stride from 16 to 32 (ablation)Axial-ResNet-L · Cityscapes valPQ (%): 63.9 -> 63.9No clear effect
replace dual convolutional decoders with axial-decoder (ablation)Axial-ResNet-L OS16 · Cityscapes valPQ (%): 63.9 -> 63.7No clear effect
use multi-scale inferenceAxial-DeepLab-L · COCO valPQ (%): 43.4 -> 43.9Improved

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