Research map / Label-efficient and robust learning / Domain adaptation and generalization
CDS: Cross-Domain Self-Supervised Pre-Training
Donghyun Kim, Kuniaki Saito, Tae-Hyun Oh, Bryan A. Plummer, Stan Sclaroff, Kate Saenko
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add cross-domain matching (replace adversarial domain classifier) (ablation) | In-domain ID · CUB universal DA, open set (DANCE) | mean class accuracy: 38.1 -> 55.9 | Improved |
| add cross-domain matching (replace adversarial domain classifier) (ablation) | In-domain ID · CUB universal DA, closed set (DANCE) | accuracy: 56.0 -> 59.0 | Improved |
| add cross-domain matching on top of in-domain instance discrimination (ablation) | In-domain ID · Office D->A | kNN accuracy: 61.7 -> 68.5 | Improved |
| add cross-domain matching on top of in-domain instance discrimination (ablation) | In-domain ID · Office D->A, CDAN 1-shot | accuracy: 49.7 -> 66.8 | Improved |
| use in-domain instance discrimination instead of naive instance discrimination (ablation) | ID · Office D->A | kNN accuracy: 59.6 -> 61.7 | Improved |
| use in-domain instance discrimination instead of naive instance discrimination (ablation) | ID · Office-Home cross-domain retrieval | P@1: 42.2 -> 44.2 | Improved |
| apply CDS as second pre-training stage after ImageNet | ImageNet-pretrained ResNet-50 · CUB cross-domain retrieval | P@1: 22.6 -> 40.9 | Improved |
| apply CDS as second pre-training stage after ImageNet | ImageNet-pretrained ResNet-50 · Office-Home cross-domain retrieval | P@1: 49.9 -> 56.3 | Improved |
| apply CDS pre-training | ImageNet init + DANCE · Office-Home universal DA, partial | accuracy: 71.1 -> 69.7 | Hurt |
| apply generic SSL (ID, SimCLR, MoCo) as second stage instead of CDS | ImageNet init · Office-Home retrieval | P@1: 49.9 -> 42.2 (ID), 48.0 (SimCLR) | Hurt |
| apply CDS pre-training to DANCE | DANCE · Office-Home open set DA | H-score: 46.6 -> 68.6 | Improved |
| apply CDS pre-training | source-only · CUB 1-shot DA | target accuracy: 5.1 -> 20.8 | Improved |
Closest papers by meaning
- Unsupervised Domain Generalization by Learning a Bridge Across Domains (CVPR 2022 · similarity 0.89)
- Prototypical Cross-Domain Self-Supervised Learning for Few-Shot Unsupervised Domain Adaptation (CVPR 2021 · similarity 0.89)
- Understanding Cross-Domain Few-Shot Learning Based on Domain Similarity and Few-Shot Difficulty (NeurIPS 2022 · similarity 0.89)
- A Broad Study of Pre-training for Domain Generalization and Adaptation (ECCV 2022 · similarity 0.88)
- Self-training For Few-shot Transfer Across Extreme Task Differences (ICLR 2021 · similarity 0.87)
- Mind the Gap Between Prototypes and Images in Cross-domain Finetuning (NeurIPS 2024 · similarity 0.86)
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.
