AI Paper Atlas

Research map / Label-efficient and robust learning / Semi-supervised learning

Semi-Supervised Semantic Segmentation With High- and Low-Level Consistency

Sudhanshu Mittal, Maxim Tatarchenko, Thomas Brox

TPAMI 2019 · Label-efficient and robust learning › Semi-supervised learning · approach: Consistency training for semi-supervised segmentation · 494 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 s4GAN branch with unlabeled images (ablation)DeepLabv2 · PASCAL VOC 5% labeledmIoU: 56.8 -> 60.9Improved
add MLMT multi-label classification branch to s4GAN (ablation)s4GAN · PASCAL VOC 5% labeledmIoU: 60.9 -> 62.9Improved
add MLMT branch only without s4GAN (ablation)DeepLabv2 · PASCAL VOC 5% labeledmIoU: 56.8 -> 59.0Improved
use mean-teacher semi-supervised classifier instead of supervised CNN classifier (ablation)s4GAN + classifier · PASCAL VOC 5% labeledmIoU: 62.2 -> 62.9Improved
use learned classifier instead of pixel-count threshold heuristic (ablation)s4GAN · PASCAL VOC 5% labeledmIoU: 61.5 -> 62.9Improved
use feature matching loss instead of standard GAN loss (ablation)CE generator loss · PASCAL VOC 1/50mIoU: 54.0 -> 55.4Improved
add self-training loss (ablation)CE + FM · PASCAL VOC 1/50mIoU: 55.4 -> 58.1Improved
use single-stage joint GAN training instead of Hung et al. two-stages4GAN · PASCAL VOC 1/50 (COCO pre-trained)mIoU: 57.2 -> 63.3Improved
add MLMT branchs4GAN · CityscapesmIoUNo clear effect
add s4GAN branchDeepLabv2 · Cityscapes 1/8 labeledmIoU: 56.2 -> 59.3Improved
use all labeled data with s4GAN onlys4GAN · PASCAL VOC full labelsmIoU: 70.7 -> 71.2No clear effect
swap backbone to DeepLabv3+s4GAN + MLMT · PASCAL VOC 1/8mIoU: 67.3 -> 70.4Improved

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