Attributes-Guided and Pure-Visual Attention Alignment for Few-Shot Recognition
Authors: Siteng Huang, Min Zhang, Yachen Kang, Donglin Wang7840-7847
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments and analysis show that our proposed module can significantly improve simple metric-based approaches to achieve state-of-the-art performance on different datasets and settings. We use two datasets with high-quality attribute annotations to conduct experiments: Caltech-UCSD-Birds 200-2011 (CUB) (Wah et al. 2011) and SUN Attribute Database (SUN) (Patterson et al. 2014). Table 1 shows the gains obtained by incorporating AGAM into each approach on two datasets, and for all three approaches, incorporating AGAM leads to a significant improvement. Ablation Study To empirically show the effectiveness of our framework design, a careful ablation study is conducted. |
| Researcher Affiliation | Academia | Siteng Huang,1,2 Min Zhang,2 Yachen Kang,2 Donglin Wang2 1 Zhejiang University 2 Machine Intelligence Lab (Mi LAB), AI Division, School of Engineering, Westlake University {huangsiteng, zhangmin, kangyachen, wangdonglin}@westlake.edu.cn |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at https://github.com/bighuang624/AGAM. |
| Open Datasets | Yes | We use two datasets with high-quality attribute annotations to conduct experiments: Caltech-UCSD-Birds 200-2011 (CUB) (Wah et al. 2011) and SUN Attribute Database (SUN) (Patterson et al. 2014). |
| Dataset Splits | No | The paper describes episodic training where support and query sets are used within each episode, and reports average accuracy over 10,000 episodes sampled from the test set. However, it does not explicitly provide global dataset splits (e.g., specific percentages or counts) for a separate validation set, nor does it specify cross-validation setup in the traditional sense. |
| Hardware Specification | No | The paper does not specify any particular hardware (e.g., GPU models, CPU types) used for the experiments. |
| Software Dependencies | No | The paper mentions using the Adam optimizer and applying standard data augmentation. However, it does not provide specific version numbers for any software dependencies like programming languages (e.g., Python), libraries (e.g., PyTorch), or frameworks. |
| Experiment Setup | Yes | Implementation Details. Our method is trained from scratch and uses the Adam (Kingma and Ba 2015) optimizer with an initial learning rate 10 3. Following the settings of (Chen et al. 2019a), we apply standard data augmentation including random crop, left-right flip, and color jitter in the meta-training stage. And for meta-learning methods, we train 60,000 episodes for 1-shot and 40,000 episodes for 5-shot settings. For AGAM, we set trade-off hyperparameters α = 1.0 and β = 0.1 for all experiments. |