Research map / Label-efficient and robust learning / Recognition and long-tail
Higher-Order Integration of Hierarchical Convolutional Activations for Fine-Grained Visual Categorization
Sijia Cai, Wangmeng Zuo, Lei Zhang
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 |
|---|---|---|---|
| use degree-2 polynomial kernel instead of linear kernel (ablation) | relu5_3 predictor · CUB (fine-tuned) | accuracy (%): 79.2 -> 83.7 | Improved |
| raise polynomial degree from 2 to 6 (ablation) | relu5_3 predictor · CUB (fine-tuned) | accuracy (%): 83.7 -> 79.5 | Hurt |
| raise polynomial degree from 2 to 3 (ablation) | relu5_3 predictor · CUB (fine-tuned) | accuracy (%): 83.7 -> 83.3 | Hurt |
| add fine-tuning (ablation) | degree-2 relu5_3 predictor · CUB | accuracy (%): 78.3 -> 83.7 | Improved |
| add relu5_2 activations (ablation) | degree-2 relu5_3 predictor · CUB (fine-tuned) | accuracy (%): 83.7 -> 85.3 | Improved |
| add relu5_1 activations to relu5_3 (ablation) | degree-2 relu5_3 predictor · CUB (fine-tuned) | accuracy (%): 83.7 -> 83.6 | No clear effect |
| add relu5_1 to relu5_3+relu5_2 (ablation) | degree-2 two-layer integration · CUB (fine-tuned) | accuracy (%): 85.3 -> 85.1 | No clear effect |
| use degree-2 kernel instead of degree-1 in two-layer integration (ablation) | relu5_3+relu5_2 · CUB (fine-tuned) | accuracy (%): 80.4 -> 85.3 | Improved |
| increase number of 1x1 filters Dr from 8,192 to 32,768 (ablation) | polynomial module · CUB | accuracy (%) | No clear effect |
| replace hypercolumn-style concatenation with max pooling by degree-2 integration (ablation) | three-layer integration · CUB | accuracy (%): 80.9 -> 85.1 | Improved |
| replace HED-style prediction fusion with degree-2 integration (ablation) | three-layer integration · CUB | accuracy (%): 82.3 -> 85.1 | Improved |
| raise polynomial degree from 2 to 6 (ablation) | polynomial module · CUB testing speed | FPS: 29.8 -> 10.4 | Hurt |
Closest papers by meaning
- Part-Stacked CNN for Fine-Grained Visual Categorization (CVPR 2016 · similarity 0.89)
- Democratizing Fine-grained Visual Recognition with Large Language Models (ICLR 2024 · similarity 0.86)
- Fine-R1: Make Multi-modal LLMs Excel in Fine-Grained Visual Recognition by Chain-of-Thought Reasoning (ICLR 2026 · similarity 0.86)
- L-WISE: Boosting Human Visual Category Learning Through Model-Based Image Selection and Enhancement (ICLR 2025 · similarity 0.85)
- Revisiting the Integration of Convolution and Attention for Vision Backbone (NeurIPS 2024 · similarity 0.85)
- Advancing Fine-Grained Classification by Structure and Subject Preserving Augmentation (NeurIPS 2024 · 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.
