Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Weak-shot Fine-grained Classification via Similarity Transfer
Junjie Chen, Li Niu, Liu Liu, Liqing Zhang
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 |
|---|---|---|---|
| Use SimTrans instead of the strongest listed baseline, Fusion2 | Weak-shot fine-grained classification · Car | accuracy: 86.9 -> 89.8 | Improved |
| Use SimTrans instead of the strongest listed baseline, Fusion2 | Weak-shot fine-grained classification · CUB | accuracy: 88.5 -> 91.7 | Improved |
| Use SimTrans instead of the strongest listed baseline, Fusion2 | Weak-shot fine-grained classification · FGVC | accuracy: 90.0 -> 92.8 | Improved |
| Add sample weighting to classification loss (ablation) | CUB classifier · CUB | accuracy: 85.4 -> 90.3 | Improved |
| Add graph regularization to classification loss (ablation) | CUB classifier · CUB | accuracy: 85.4 -> 89.5 | Improved |
| Add sample weighting and graph regularization (ablation) | CUB classifier · CUB | accuracy: 85.4 -> 91.2 | Improved |
| Add adversarial loss to SimNet while using sample weighting (ablation) | CUB classifier · CUB | accuracy: 90.3 -> 90.7 | Improved |
| Add adversarial loss to SimNet while using graph regularization (ablation) | CUB classifier · CUB | accuracy: 89.5 -> 90.1 | Improved |
| Use the full method rather than classification loss alone (ablation) | CUB classifier · CUB | accuracy: 85.4 -> 91.7 | Improved |
| Support the classifier with SimNet similarity learned from clean base categories rather than noisy novel data (ablation) | CUB classifier · CUB | accuracy: 87.6 -> 91.7 | Improved |
| Use end-to-end SimNet similarity rather than Euclidean similarity from base data (ablation) | CUB classifier · CUB | accuracy: 88.8 -> 91.7 | Improved |
| Reduce base training set to 50 categories with 5 images per category (ablation) | SimTrans classifier · CUB | accuracy: 91.7 -> 89.3 | Hurt |
Closest papers by meaning
- Webly Supervised Learning Meets Zero-Shot Learning: A Hybrid Approach for Fine-Grained Classification (CVPR 2018 · similarity 0.85)
- Learning from Web Data Using Adversarial Discriminative Neural Networks for Fine-Grained Classification (AAAI 2019 · similarity 0.85)
- CaT: Weakly Supervised Object Detection With Category Transfer (ICCV 2021 · similarity 0.84)
- Diversity Transfer Network for Few-Shot Learning (AAAI 2020 · similarity 0.84)
- ConFeSS: A Framework for Single Source Cross-Domain Few-Shot Learning (ICLR 2022 · similarity 0.84)
- Adversarially Robust Few-shot Learning via Parameter Co-distillation of Similarity and Class Concept Learners (CVPR 2024 · similarity 0.83)
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.
