Reproduce “Few-shot Fine-grained Image Classification via Multi-Frequency Neighborhood and Double-cross Modulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hegui Zhu, Zhan Gao, Jiayi Wang, Yange Zhou, Chengqing Li
FicNet enhances few-shot fine-grained image classification by extracting compact multi-frequency structural representations and adaptively modulating features with global context and task-specific attention. The model achieved 93.17% accuracy on CUB-200 and 95.36% on Stanford Cars for 5-shot tasks, outperforming existing approaches.
Key insights
- Stable structural representations, derived from multi-frequency components and weighted neighborhoods, minimize intra-class distances by filtering out irrelevant variations.
- Incorporating global contextual information via crisscross operations provides a comprehensive understanding of the target object, enhancing feature completeness.
- A double 3D cross-attention mechanism adaptively aligns discriminative features between support and query samples, focusing on task-relevant subtle cues.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.