Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Meta-Reinforced Synthetic Data for One-Shot Fine-Grained Visual Recognition
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 |
|---|---|---|---|
| Learn to fuse real and GAN-generated images with MetaIRNet | ProtoNet one-shot classification · CUB 5-way-1-shot | accuracy (%): 81.73% vs. 84.13% | Helped |
| Learn to fuse real and GAN-generated images with MetaIRNet | ProtoNet one-shot classification · NAB 5-way-1-shot | accuracy (%): 87.91% vs. 89.19% | Helped |
| Add GAN-generated images directly without image fusion (ablation) | ProtoNet · CUB 5-way-1-shot | accuracy (%): 81.73 -> 79.40 | Hurt |
| Add GAN-generated images directly without image fusion (ablation) | ProtoNet · NAB 5-way-1-shot | accuracy (%): 87.91 -> 85.40 | Hurt |
| Add horizontal flip augmentation (ablation) | ProtoNet · CUB 5-way-1-shot | accuracy (%): 81.73 -> 82.66 | Helped |
| Add random flip augmentation to MetaIRNet (ablation) | MetaIRNet + FinetuneGAN · CUB 5-way-1-shot | accuracy (%): 84.13% to 84.80% | Helped |
| Increase generated support augmentations from 1 to 2 (ablation) | MetaIRNet · CUB 5-way-1-shot | accuracy (%): 84.13 ± 0.60 -> 83.45 ± 0.60 | Hurt |
| Increase generated support augmentations from 1 to 3, 5, or 10 (ablation) | MetaIRNet · CUB 5-way-1-shot | accuracy (%): 84.13 ± 0.60; 80.99 ± 0.62; 81.21 ± 0.62; 80.49 ± 0.69 | Hurt |
| Use random 1×1 image mixing instead of learned grid-based fusion (ablation) | MetaIRNet image fusion · CUB 5-way-1-shot | accuracy (%): 82.24 ± 0.59 vs. 84.13 ± 0.58 | Hurt |
| Use random real gallery images instead of GAN-synthesized images for fusion (ablation) | MetaIRNet image fusion · CUB 5-way-1-shot | accuracy (%): 82.84 ± 0.62 vs. 84.13 ± 0.58 | Hurt |
| Use random real gallery images instead of GAN-synthesized images for fusion (ablation) | MetaIRNet image fusion · NAB 5-way-1-shot | accuracy (%): 88.42 ± 0.59 vs. 89.19±0.51 | Hurt |
| Train with the simplified generator from scratch without classifier-to-generator gradients (ablation) | MetaIRNet · CUB 5-way-1-shot | accuracy (%): 63.77 ± 0.71 | Mixed |
Closest papers by meaning
- On sensitivity of meta-learning to support data (NeurIPS 2021 · similarity 0.87)
- Partial Is Better Than All: Revisiting Fine-tuning Strategy for Few-shot Learning (AAAI 2021 · similarity 0.86)
- Bi-directional Feature Reconstruction Network for Fine-Grained Few-Shot Image Classification (AAAI 2023 · similarity 0.86)
- Neural Fine-Tuning Search for Few-Shot Learning (ICLR 2024 · similarity 0.85)
- Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024 · similarity 0.85)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 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.