AI Paper Atlas

Research map / Label-efficient and robust learning / Domain adaptation and generalization

Effective Use of Synthetic Data for Urban Scene Semantic Segmentation

Fatemeh Sadat Saleh, Mohammad Sadegh Aliakbarian, Mathieu Salzmann, Lars Petersson, Jose M. Alvarez

ECCV 2018 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Domain-generalized segmentation · 164 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 detection-based (Mask R-CNN) handling of foreground classes to DeepLab trained on GTA5+VEISDeepLab trained on synthetic data · Cityscapes valmIoU: 32.8 -> 38.0Improved
add VEIS data to GTA5 training in a plain semantic segmentation networkDeepLab · Cityscapes valmIoU: 31.3 -> 32.8Improved
add pseudo-labels on unlabeled real images to the plain GTA5+VEIS baselineGTA5+VEIS DeepLab · Cityscapes valmIoU: 32.8 -> 33.3Improved
add pseudo-labels from the detection-based method on unlabeled real imagesOurs · Cityscapes valmIoU: 38.0 -> 42.5Improved
add pseudo-labels from the detection-based method on unlabeled real imagesOurs · CamVid testmIoU: 47.6 -> 48.8Improved
add detection-based foreground handling with VEIS to GTA5 baselineGTA5 DeepLab · CamVid testmIoU: 44.6 -> 47.6Improved
add VEIS data to GTA5 in a plain segmentation networkGTA5 DeepLab · CamVid testmIoU: 44.6 -> 44.4No clear effect
replace GTA5 by VEIS as training set for a plain segmentation networkDeepLab · Cityscapes valmIoU: 31.3 -> 24.4Hurt
apply the detection-based approach when training on Cityscapes real labeled datafully supervised DeepLab · Cityscapes valmIoU: 56.2 -> 60.0Improved
use texture instead of silhouette as classifier input for real vs synthetic discriminationbinary VGG-16 classifier · VEIS vs Cityscapes objectsaccuracy: 70.0% -> 95.1%Improved
train object classifier on textured silhouettes instead of plain silhouettesmulti-class VGG-16 classifier · Cityscapes real objects testaccuracy gap real vs synthetic: 81.0 vs 89.2 (silhouette); 83.7 vs 94.2 (textured)Hurt

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