Research map / Label-efficient and robust learning / Recognition and long-tail
FD²: A Dedicated Framework for Fine-Grained Dataset Distillation
Hongxu Ma, Guang Li, Shijie Wang, DONGZHAN ZHOU, Baoli Sun, Takahiro Ogawa, Miki Haseyama, Zhihui Wang
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 fine-grained characteristic constraint (ablation) | SRe2L++ · CUB-200-2011 IPC=3 ResNet18 | Top-1 accuracy: 63.4 -> 64.8 | Improved |
| add similarity constraint (ablation) | SRe2L++ · CUB-200-2011 IPC=3 ResNet18 | Top-1 accuracy: 63.4 -> 64.6 | Improved |
| add both constraints together (ablation) | SRe2L++ · CUB-200-2011 IPC=3 ResNet18 | Top-1 accuracy: 63.4 -> 64.9 | Improved |
| increase group size N_S from 4 to 5 in similarity constraint (ablation) | SRe2L++_FD2 · CUB-200-2011 IPC=5 | Top-1 accuracy: 66.1 -> 64.6 | Hurt |
| set beta to 0.5 instead of 0 in fine-grained constraint (ablation) | SRe2L++_FD2 · CUB-200-2011 IPC=3 | Top-1 accuracy: 63.5 -> 64.8 | Improved |
| set CAL ratio alpha to 0.5 instead of 0.7 (ablation) | SRe2L++_FD2 · CUB-200-2011 IPC=3 | Top-1 accuracy: 62.1 -> 63.4 | Improved |
| set constraint weight lambda to 0.8 instead of 0.2 | SRe2L++_FD2 · CUB-200-2011 | Top-1 accuracy: 65.4 -> 67.0 | Improved |
| integrate FD2 into FADRM+ | FADRM+ · Stanford Cars ResNet18 IPC=1 | Top-1 accuracy: 60.3 -> 74.1 | Improved |
| integrate FD2 into FADRM+ | FADRM+ · CUB-200-2011 ResNet50 IPC=3 | Top-1 accuracy: 70.4 -> 70.6 | No clear effect |
| integrate FD2 into FADRM+ with small student | FADRM+ · ImageNette ResNet18 IPC=1 | Top-1 accuracy: 39.2 -> 38.6 | Hurt |
| integrate FD2 into FADRM+ with larger student | FADRM+ · ImageNette ResNet50 IPC=1 | Top-1 accuracy: 31.9 -> 39.0 | Improved |
| integrate FD2 into FADRM+ | FADRM+ · ImageWoof ResNet50 IPC=1 | Top-1 accuracy: 19.9 -> 30.3 | Improved |
Closest papers by meaning
- Enhancing Dataset Distillation via Non-Critical Region Refinement (CVPR 2025 · similarity 0.89)
- Breaking Class Barriers: Efficient Dataset Distillation via Inter-Class Feature Compensator (ICLR 2025 · similarity 0.89)
- Fine-grained Classes and How to Find Them (ICML 2024 · similarity 0.89)
- Rethinking Dataset Distillation for Classification: Do Distilled Sets Outperform Coresets? (NeurIPS 2026 · similarity 0.88)
- A Label is Worth A Thousand Images in Dataset Distillation (NeurIPS 2024 · similarity 0.88)
- Boost Self-Supervised Dataset Distillation via Parameterization, Predefined Augmentation, and Approximation (ICLR 2025 · 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.
