Research map / Label-efficient and robust learning / Recognition and long-tail
Learning from Web Data Using Adversarial Discriminative Neural Networks for Fine-Grained Classification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 web data to clean training data without adversarial loss (ablation) | ResNet50 fine-tuning · Food-101 | accuracy (%): 84.31 -> 85.21 | Helped |
| add adversarial discriminative source loss on top of mixed web+clean training (ablation) | ResNet50 · Food-101 | accuracy (%): 85.21 -> 89.35 | Helped |
| add adversarial discriminative source loss on top of mixed training (ablation) | ResNet50 · MIT Indoor 67 | accuracy (%): 82.35 -> 84.59 | Helped |
| add adversarial discriminative source loss on top of mixed training (ablation) | VggNet · Stanford Dogs | accuracy (%): 81.03 -> 84.92 | Helped |
| filter noisy web images by predicted label before training (ablation) | CaffeNet · MIT Indoor 67 | accuracy (%): 68.00 -> 63.53 | Hurt |
| filter noisy web images by predicted label before training (ablation) | VggNet · Stanford Dogs | accuracy (%): 81.03 -> 79.70 | Hurt |
| filter noisy web images by predicted label before training (ablation) | ResNet50 · Food-101 | accuracy (%): 85.21 -> 86.10 | Helped |
| set adversarial trade-off lambda from 0 to 0.1 (ablation) | adversarial multi-task model · Food-101, Stanford Dogs | validation accuracy | Helped |
| increase lambda above 0.3 (ablation) | adversarial multi-task model · MIT Indoor 67 | validation accuracy | Hurt |
| proposed method vs Goldfince (multi-crop, extra categories) | ResNet50 · Stanford Dogs | accuracy (%): 85.90 -> 87.07 | Helped |
Closest papers by meaning
- Web-Supervised Network with Softly Update-Drop Training for Fine-Grained Visual Classification (AAAI 2020 · similarity 0.87)
- Progressive Adversarial Networks for Fine-Grained Domain Adaptation (CVPR 2020 · similarity 0.87)
- Fine-Grained Recognition: Accounting for Subtle Differences between Similar Classes (AAAI 2020 · similarity 0.86)
- A Compact DNN: Approaching GoogLeNet-Level Accuracy of Classification and Domain Adaptation (CVPR 2017 · similarity 0.85)
- Adversarial Teacher-Student Representation Learning for Domain Generalization (NeurIPS 2021 · similarity 0.85)
- Attend in Groups: A Weakly-Supervised Deep Learning Framework for Learning From Web Data (CVPR 2017 · similarity 0.85)
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.