Research map / Label-efficient and robust learning / Domain adaptation and generalization
Grounding Visual Representations with Texts for Domain Generalization
Seonwoo Min, Nokyung Park, Siwon Kim, Seunghyun Park, Jinkyu Kim
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 Visual and Textual Joint Embedder (ablation) | Ours w/ PTE · CUB-DG multi-source (avg over 4 target domains) | accuracy: 49.3 -> 57.0 | Improved |
| add Textual Explanation Generator (ablation) | Ours w/ PTE · CUB-DG multi-source (avg over 4 target domains) | accuracy: 56.1 -> 57.0 | Improved |
| use per-image texts instead of per-class texts (ablation) | Ours w/ PTE · CUB-DG multi-source (avg over 4 target domains) | accuracy: 56.5 -> 57.0 | Improved |
| raise lambda_align from 0.1 to 1.0 (lambda_expl 1.0) (ablation) | Ours w/ PTE · CUB-DG multi-source (avg over 4 target domains) | accuracy: 54.8 -> 57.0 | Improved |
| raise lambda_expl from 0.1 to 1.0 (lambda_align 1.0) (ablation) | Ours w/ PTE · CUB-DG multi-source (avg over 4 target domains) | accuracy: 55.9 -> 57.0 | Improved |
| raise both lambdas from 0.1 to 1.0 (ablation) | Ours w/ PTE · CUB-DG multi-source (avg over 4 target domains) | accuracy: 53.5 -> 57.0 | Improved |
| replace CLIP text encoder with MPNet, DistillBERT or MiniLM (ablation) | Ours w/ PTE · CUB-DG multi-source (avg over 4 target domains) | accuracy: 57.0 -> 56.3 | No clear effect |
| use self-supervised textual encoder instead of pre-trained CLIP encoder | Ours · CUB-DG multi-source (avg over 4 target domains) | accuracy: 57.0 vs 56.6 | No clear effect |
| add joint embedder to explanation generator (ablation) | Ours · Generated explanation quality (CUB-DG) | CIDEr-D: 28.4 -> 40.7 | Improved |
| add joint embedder to explanation generator (ablation) | Ours · Generated explanation quality (CUB-DG) | BLEU-4: 42.9 -> 48.0 | Improved |
| add text grounding vs best baseline SD | Ours w/ PTE · CUB-DG single-source (avg over target domains) | accuracy: 41.7 -> 45.7 | Improved |
| add text grounding vs best baseline CORAL | Ours w/ PTE · CUB-DG multi-source (avg over 4 target domains) | accuracy: 55.4 -> 57.0 | Improved |
Closest papers by meaning
- Generalizing from SIMPLE to HARD Visual Reasoning: Can We Mitigate Modality Imbalance in VLMs? (ICML 2025 · similarity 0.89)
- Leveraging Vision-Language Models for Improving Domain Generalization in Image Classification (CVPR 2024 · similarity 0.89)
- Reasoning-Driven Multimodal LLM for Domain Generalization (ICLR 2026 · similarity 0.89)
- Language Model as Visual Explainer (NeurIPS 2024 · similarity 0.88)
- Explaining Domain Shifts in Language: Concept Erasing for Interpretable Image Classification (CVPR 2025 · similarity 0.88)
- Bridging Vision and Language Spaces with Assignment Prediction (ICLR 2024 · similarity 0.88)
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.
