AI Paper Atlas

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

Tackling Long-Tailed Category Distribution under Domain Shifts

Xiao Gu, Yao Guo, Zeju Li, Jianing Qiu, Qi Dou, Yuxuan Liu, Benny Lo, Guang-Zhong Yang

ECCV 2022 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Optimization-based domain generalization · 14 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
use the full meta-learning method instead of aggregate training (Agg)ResNet-10 · AWA2-LTS (held-out Original)Acc-U: 29.4 -> 49.4Improved
use the full method instead of aggregate training (Agg)ResNet-10 · ImageNet-LTS (held-out Original)Acc-U: 19.5 -> 24.3Improved
use full method instead of the best combination MixStyle+BSCEResNet-10 · AWA2-LTS (held-out Original)Acc-U: 40.0 -> 49.4Improved
add warm-up pretraining to DAMLDAML · AWA2-LTS (held-out Original)Acc-U: 13.9 -> 42.2Improved
add distribution-calibrated loss in place of BSCE (ablation)long-tail loss · AWA2-LTS / ImageNet-LTSAcc-UImproved
use domain-specific prototypes for cross-domain alignment instead of a uni-domain prototype (ablation)visual-semantic mapping · AWA2-LTSAcc-UImproved
use the n_k weighted term in covariance update (ablation)augmentation module · AWA2-LTSAcc-UImproved
change CLIP or GloVe embeddings instead of BERTOurs · AWA2-LTSH: 44.6 -> 43.6 (CLIP), 44.5 (GloVe)No clear effect
use Ours instead of Agg at imbalance ratio 10ResNet-10 · AWA2-LTS (imbalance ratio 10)H: 20.7 -> 32.1Improved
use Ours instead of Agg at imbalance ratio 50ResNet-10 · AWA2-LTS (imbalance ratio 50)H: 32.8 -> 41.0Improved
use Ours instead of AggResNet-18 pretrained · PACS-ODG (average)Acc-U: 55.2 -> 67.1Improved
use Ours instead of DAMLResNet-18 pretrained · PACS-ODG (average)H-U: 51.9 -> 56.9Improved

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