AI Paper Atlas

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

Contrastive Model Adaptation for Cross-Condition Robustness in Semantic Segmentation

David Brüggemann, Christos Sakaridis, Tim Broedermann, Luc Van Gool

ICCV 2023 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Adaptation for semantic segmentation · 23 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 cross-domain contrastive (CDC) loss with reference images (ablation)CMA · ACDC valmIoU: 60.1 -> 67.2Improved
global average pooled embeddings in contrastive loss (ablation)CMA · ACDC valmIoU: +4.6% mIoUImproved
patch-level grouping on 7x7 grid with reference warping instead of global embeddings (ablation)CMA · ACDC valmIoU: +1.2% mIoUImproved
use patches without spatial warpingCMA · ACDC valmIoUHurt
add confidence modulation to patch forming (ablation)CMA · ACDC valmIoU: +1% mIoUImproved
add reference images naively (mixed or curriculum) instead of contrastive loss (ablation)URMA / CMA w/o CDC · ACDC valmIoU: 63.2 -> 62.9 (URMA mixed); 60.1 -> 60.3 (CMA mixed)No clear effect
use contrastive reference-image loss vs best naive strategyCMA · ACDC valmIoU: 64.1 -> 67.2Improved
swap InfoNCE for RINCE or debiased loss (ablation)CMA · ACDC valmIoU: 67.2 -> 67.4 (RINCE); 66.3 (debiased)No clear effect
add self-training loss (ablation)CMA · ACDC valmIoU: 57.7 -> 67.2Improved
add entropy loss (ablation)CMA · ACDC valmIoU: 66.7 -> 67.2Improved
vary grid size, temperature, confidence threshold (ablation)CMA · ACDC valmIoU: 66.7 to 67.2 across thresholdsNo clear effect
adapt with CMA instead of source model on DeepLabv2 vs SegFormerCMA · Cityscapes to ACDC testmIoU: SegFormer 59.4 -> 69.1; DeepLabv2 39.5 -> 50.4Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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