AI Paper Atlas

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

ADeLA: Automatic Dense Labeling With Attention for Viewpoint Shift in Semantic Segmentation

Hanxiang Ren, Yanchao Yang, He Wang, Bokui Shen, Qingnan Fan, Youyi Zheng, C. Karen Liu, Leonidas J. Guibas

CVPR 2022 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Domain-generalized segmentation · 2 citations (OpenAlex)

Check 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
Use attention-based spatial transportation instead of UNet color transformation for label hallucination (ablation)ADeLA · Target domains 10°–90°mIoU of hallucinated labelsImproved
Add functional label hallucination instead of nearest-neighbor conversion (ablation)ADeLA (single) · Target domains 10°–90°mIoU of hallucinated labels: 54.85 -> 70.07; 46.29 -> 67.63; 42.66 -> 58.62; 37.75 -> 54.33; 27.71 -> 47.45; 21.33 -> 37.81; 14.18 -> 28.39; 8.69 -> Improved
Use single-target rather than multiple-target view hallucination (ablation)ADeLA · Target domains 10°–70°mIoU of hallucinated labels: 48.42 -> 54.85; 41.87 -> 46.29; 35.73 -> 42.66; 30.39 -> 37.75; 24.11 -> 27.71; 17.40 -> 21.33; 11.79 -> 14.18Improved
Use multiple-target rather than single-target view hallucination (ablation)ADeLA · 80° and 90° target domainsmIoU of hallucinated labels: 8.69 -> 8.82; 4.17 -> 7.34Mixed
Use learned flow-based 2D correspondence rather than UNet transformation (ablation)View hallucination · Large viewpoint changesmIoU of hallucinated labelsImproved
Use learned flow-based correspondence rather than fully supervised RAFT correspondence (ablation)View hallucination · Small viewpoint changesmIoU of hallucinated labelsHurt
Use fully supervised RAFT correspondence rather than learned Flow (ablation)View hallucination · Large viewpoint changesmIoU of hallucinated labelsHurt
Use 3D view synthesis rather than UNet transformation (ablation)View hallucination · Target domainsmIoU of hallucinated labelsHurt
Add functional label hallucination (ablation)UNet · Target domains 10°–90°mIoU of hallucinated labels: 49.76 -> 73.62; 28.19 -> 49.07; 13.69 -> 27.12; 9.26 -> 20.08; 6.56 -> 16.48; 4.71 -> 13.68; 2.59 -> 11.61; 1.63 -> 9.79Improved
Use ADeLA instead of the strongest listed baselineSemantic segmentation adaptation · 10° target domain testmIoU: 37.80 -> 31.19Hurt
Use ADeLA instead of the strongest listed baselineSemantic segmentation adaptation · 20°–90° target domain testsmIoU: 26.74 -> 27.66; 24.15 -> 24.31; 17.75 -> 20.71; 8.66 -> 15.88; 5.68 -> 12.28; 4.67 -> 8.27; 3.87 -> 7.16; 3.39 -> 5.03Improved

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