Research map / Label-efficient and robust learning / Recognition and long-tail
Adversarial Reconstruction Feedback for Robust Fine-grained Generalization
Shijie Wang, Jian Shi, Haojie Li
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 adversarial reconstruction feedback to a classification-only retrieval model (ablation) | ResNet-50 retrieval · CUB-200-2011 | Recall@1: 66.3% -> 76.8% | Improved |
| use reconstruction feedback instead of classification feedback for retrieval (ablation) | retrieval model · CUB-200-2011 | Recall@1: 66.3% -> 62.2% | Hurt |
| add both discrepancy and non-discrepancy reconstruction losses (ablation) | AdvRF · CUB-200-2011 | Recall@1: 73.7% -> 76.8% | Improved |
| add parameterization (distillation) loss L_P (ablation) | AdvRF · CUB-200-2011 | Recall@1: 76.8% -> 76.6% | No clear effect |
| adversarial reconstruction versus non-adversarial reconstruction (ablation) | AdvRF · CUB-200-2011 | Recall@1: 72.6% -> 76.6% | Improved |
| learned discrepancy decoupling instead of CAM localization (ablation) | AdvRF · CUB-200-2011 | Recall@1: 69.8% -> 76.6% | Improved |
| learned discrepancy decoupling instead of bounding box localization (ablation) | AdvRF · CUB-200-2011 | Recall@1: 73.9% -> 76.6% | Improved |
| increase mean-generator updating ratio from 0.2 to 0.8 (ablation) | AdvRF · CUB-200-2011 | Recall@1: 76.6% -> 73.9% | Hurt |
| lower mean-generator updating ratio from 0.2 to 0.1 (ablation) | AdvRF · CUB-200-2011 | Recall@1: 76.6% -> 75.4% | Hurt |
Closest papers by meaning
- Fine-Grained Image Retrieval via Dual-Vision Adaptation (AAAI 2026 · similarity 0.88)
- Language-driven Fine-grained Retrieval (CVPR 2026 · similarity 0.87)
- Towards Optimal Fine Grained Retrieval via Decorrelated Centralized Loss with Normalize-Scale Layer (AAAI 2019 · similarity 0.87)
- Learning to Align Generative Appearance Priors for Fine-grained Image Retrieval (NeurIPS 2026 · similarity 0.86)
- LORE: Lagrangian-Optimized Robust Embeddings for Visual Encoders (NeurIPS 2025 · similarity 0.85)
- Composed Image Retrieval with Text Feedback via Multi-grained Uncertainty Regularization (ICLR 2024 · similarity 0.84)
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.
