AI Paper Atlas

Research map / Detection and segmentation / Weakly supervised detection and segmentation

Mining Cross-Image Semantics for Weakly Supervised Semantic Segmentation

Guolei Sun, Wenguan Wang, Jifeng Dai, Luc Van Gool

ECCV 2020 · Detection and segmentation › Weakly supervised detection and segmentation · approach: CAM refinement and prototypes

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 co-attention (common semantics) loss (ablation)basic classifier · PASCAL VOC12 valmIoU: 61.7 -> 65.5Improved
add contrastive co-attention loss (ablation)co-attention classifier · PASCAL VOC12 valmIoU: 65.5 -> 66.2Improved
train with co-attention and contrastive co-attention (single-round inference) (ablation)basic classifier · PASCAL VOC12 valmIoU: 61.7 -> 64.7Improved
use multi-round co-attentive inference with related images (ablation)co-attention classifier · PASCAL VOC12 valmIoU: 64.7 -> 66.2Improved
use contrastive co-attention in localization map inference (ablation)full model · PASCAL VOC12 valmIoU: 66.2 -> 66.2No clear effect
increase related images from 0 to 3 (ablation)co-attention inference · PASCAL VOC12 valmIoU: 64.7 -> 66.2Improved
increase related images from 3 to 4 (ablation)co-attention inference · PASCAL VOC12 valmIoU: 66.2 -> 66.1Hurt
increase related images from 3 to 5 (ablation)co-attention inference · PASCAL VOC12 valmIoU: 66.2 -> 66.0Hurt
use ResNet instead of VGG backboneOurs · PASCAL VOC12 val (VOC only)mIoU: 63.5 -> 66.2Improved
add extra single-label imagesOurs (ResNet) · PASCAL VOC12 valmIoU: 66.2 -> 67.1Improved
add extra web imagesOurs (ResNet) · PASCAL VOC12 valmIoU: 66.2 -> 67.7Improved
use co-attention method instead of AttnBN with extra single-label dataAttnBN · PASCAL VOC12 (extra single-label images) valmIoU: 62.1 -> 67.1Improved

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