Research map / Label-efficient and robust learning / Domain adaptation and generalization
Class-Incremental Domain Adaptation
Jogendra Nath Kundu, Rahul Mysore Venkatesh, Naveen Venkat, Ambareesh Revanur, R. Venkatesh Babu
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 trainable guides via domain projection networks (target-specific latent space) (ablation) | Ours · Office (20 shared/31 classes, one-shot) | avg accuracy, all classes (%): 66.4 -> 72.4 | Improved |
| add regularization loss L_r on proxy-source samples (ablation) | Ours · Office (20 shared/31 classes, one-shot) | avg accuracy, all classes (%): 67.5 -> 72.4 | Improved |
| add compactness loss L_a2 (ablation) | Ours · Office (20 shared/31 classes, one-shot) | avg accuracy, private classes (%): 68.1 -> 71.8 | Improved |
| use real source data instead of Gaussian prototype proxy samples (ablation) | Ours · Office (20 shared/31 classes, one-shot) | avg accuracy, all classes (%): 72.4 -> 72.9 | No clear effect |
| add class separability objective L_s1 in source training (ablation) | Ours · Office A->D (one-shot) | accuracy (%): 68.6 -> 72.2 | Improved |
| add negative training samples in source training (ablation) | Ours · Office A->D (one-shot) | accuracy (%): 67.4 -> 72.2 | Improved |
| add negative training samples in source training (ablation) | Ours · Office A->D (one-shot) | private-class accuracy (%): 62.3 -> 72.6 | Improved |
| add negative training samples in source training (ablation) | source model · Office A->D | source accuracy (%): 96.9 -> 96.7 | No clear effect |
| add trainable guides via domain projection networks (ablation) | Ours · Office (20 shared/31 classes, one-shot) | avg private-class accuracy (%): 64.5 -> 71.8 | Improved |
| two-step model upgrade (A to D to W) instead of direct adaptation | Ours · Office W target | accuracy (%): 80.3 -> 79.9 | No clear effect |
| use proposed method instead of UAN+TPC (few-shot) | UAN+TPC · Office (10/20 classes, few-shot) | avg private accuracy: 28.6 -> 75.3 | Improved |
Closest papers by meaning
- Prototype-Guided Continual Adaptation for Class-Incremental Unsupervised Domain Adaptation (ECCV 2022 · similarity 0.91)
- Active Universal Domain Adaptation (ICCV 2021 · similarity 0.89)
- Minimum Class Confusion for Versatile Domain Adaptation (ECCV 2020 · similarity 0.89)
- Associative Domain Adaptation (ICCV 2017 · similarity 0.89)
- Category-Aware Active Domain Adaptation (ICML 2024 · similarity 0.88)
- Reviewing the Forgotten Classes for Domain Adaptation of Black-Box Predictors (AAAI 2024 · similarity 0.87)
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.
