AI Paper Atlas

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

Online Domain Adaptation for Semantic Segmentation in Ever-Changing Conditions

Theodoros Panagiotakopoulos, Pier Luigi Dovesi, Linus Härenstam-Nielsen, Matteo Poggi

ECCV 2022 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Adaptation for semantic segmentation · 24 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
adapt progressively online (OnDA Hybrid Switch) instead of offline adaptation to all rain at onceDeepLabv2-ResNet50 · Cityscapes rain, 200mmmIoU: 23.7 -> 42.2Improved
add Mahalanobis (variance-normalized) distance for prototypes instead of Euclidean (ablation)OnDA · hardest target domainmIoU: +4.5%Improved
update prior to dynamic model instead of static model (ablation)OnDA · Cityscapes rain 200mm forwardmIoU: 37.8 -> 41.2Improved
use dynamic prior instead of static prior when adapting back to clear (ablation)OnDA · Cityscapes rain backward, clearmIoU: 64.8 -> 53.3Hurt
switch between static and dynamic priors (Confidence Derivative Switch) (ablation)OnDA - Dynamic Model · Cityscapes rain backward (h-mean)harmonic-mean mIoU: 50.6 -> 55.2Improved
use Hybrid Switch instead of Soft-Confidence Switch (ablation)OnDA · Cityscapes rain backward (h-mean)harmonic-mean mIoU: 52.5 -> 55.1Improved
add online adaptation (OnDA) versus TENTTENT · Cityscapes rain forward (h-mean)harmonic-mean mIoU: 35.1 -> 54.2Improved
add Replay Buffer to TENT (ablation)TENT · Cityscapes rain forward (h-mean)harmonic-mean mIoU: 35.1 -> 39.7Improved
add replay buffer of 100 source images (ablation)OnDA Hybrid Switch · Cityscapes rain clear, backwardmIoU: 57.5 -> 63.0Improved
increase replay buffer from 1000 to whole dataset (ablation)OnDA Hybrid Switch · Cityscapes rain clear, backwardmIoU: 64.8 -> 65.4No clear effect
run 3 times faster storm (One Pass) (ablation)OnDA Hybrid Switch · Cityscapes rain backward (h-mean)harmonic-mean mIoU: 55.1 -> 56.3Mixed
start from pre-adapted model before storm C (200mm first) (ablation)OnDA Hybrid Switch · Storm CmIoUImproved

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