AI Paper Atlas

Research map / Detection and segmentation / CNN features

CCNet: Criss-Cross Attention for Semantic Segmentation

Zilong Huang, Xinggang Wang, Lichao Huang, Chang Huang, Yunchao Wei, Wenyu Liu

ICCV 2019 · Detection and segmentation › CNN features · approach: Context modeling for semantic segmentation · 3,299 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 one criss-cross attention loop (R=1) (ablation)ResNet-101 FCN baseline · Cityscapes valmIoU (%): 75.1 -> 78.0Improved
increase RCCA loops from 1 to 2 (ablation)ResNet-101 RCCA · Cityscapes valmIoU (%): 78.0 -> 79.8Improved
increase RCCA loops from 2 to 3 (ablation)ResNet-101 RCCA · Cityscapes valmIoU (%): 79.8 -> 80.2Improved
add category consistent loss (CCL) (ablation)ResNet-101 RCCA(R=2) · Cityscapes valmIoU (%): 79.8 -> 80.5Improved
add category consistent loss (CCL) (ablation)ResNet-50 RCCA(R=2) · Cityscapes valmIoU (%): 78.5 -> 79.3Improved
use piece-wise instead of quadratic distance in CCL (ablation)ResNet-50 RCCA+CCL · Cityscapes valtraining success rate: 6/10 -> 9/10Improved
use piece-wise instead of quadratic distance in CCL (ablation)ResNet-50 RCCA+CCL · Cityscapes valmIoU (%): 79.2 -> 79.3No clear effect
use RCCA(R=2) instead of non-local block (ablation)ResNet-50 · Cityscapes valmIoU (%): 77.3 -> 78.5Improved
use RCCA instead of non-local in terms of memory (ablation)ResNet-50 · Cityscapes valmemory increment (M): 1411 -> 127Improved
run non-local module recurrently (R=2) (ablation)ResNet-50 +NL · Cityscapes valmIoU (%): 77.3 -> 78.7Improved
use full criss-cross shape instead of stacked horizontal-vertical attention (ablation)ResNet-50 · Cityscapes valmIoU (%): 77.3 -> 78.5Improved
increase input frames from T=1 to T=5 in 3D criss-cross attention (ablation)CCNet3D · CamVid testmIoU (%): 77.9 -> 79.1Improved

Later papers that compared with this method

16 later accepted papers report CCNet as a baseline in their results tables; 13 of them report a better number on most of the results they share with it.

8 more not listed.

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