FeatWalk: Enhancing Few-Shot Classification through Local View Leveraging
Authors: Dalong Chen, Jianjia Zhang, Wei-Shi Zheng, Ruixuan Wang
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on multiple benchmark datasets consistently demonstrate the effectiveness and versatility of our method. The source code is available at https://github.com/exceefind/ Feat Walk. |
| Researcher Affiliation | Academia | School of Computer Science and Engineering, Sun Yat-sen University, Guangzhou, China Key Laboratory of Machine Intelligence and Advanced Computing, MOE, Guangzhou, China School of Biomedical Engineering, Shenzhen Campus of Sun Yat-sen University, Guangdong, China Peng Cheng Laboratory, Shenzhen, China |
| Pseudocode | No | No pseudocode or algorithm block was found in the paper. |
| Open Source Code | Yes | The source code is available at https://github.com/exceefind/ Feat Walk. |
| Open Datasets | Yes | Mini Image Net (Vinyals et al. 2016) is a subset of the ILSVRC-12 dataset commonly used for few-shot learning, consisting of 100 classes with 600 samples per class. We used the same split as in previous studies (Ravi and Larochelle 2017) , with 64, 16, and 20 classes for training, validation, and testing, respectively. Tiered Image Net (Ren et al. 2018) is a larger dataset based on the ILSVRC-12 dataset. It consists of 34 superclasses, each containing 20 sub-classes. Following the original work (Ren et al. 2018), we used 351, 97, and 160 classes for training, validation, and test, respectively. CUB (Wah et al. 2011) is a fine-grained few-shot benchmark that includes 200 classes of birds. Following previous work (Chen et al. 2019), we used 100, 50, and 50 classes for training, validation, and test, respectively. |
| Dataset Splits | Yes | We used the same split as in previous studies (Ravi and Larochelle 2017) , with 64, 16, and 20 classes for training, validation, and testing, respectively. |
| Hardware Specification | No | The paper mentions using ResNet-12 and ResNet-18 backbones but provides no specific details about the hardware (e.g., GPU model, CPU type, memory) used for experiments. |
| Software Dependencies | No | The paper mentions using the AdamW optimizer but does not specify version numbers for any software dependencies or libraries. |
| Experiment Setup | Yes | During the training of the classifier head, we use the Adam W optimizer for simple and fast adaptation learning, with the learning rate of 1e-3, and optimize for 100 epochs. For each image, we randomly select 12 local views. Each local view is a 20% sized patch of the image. The temperature parameter τ is set to 32, and α is set to 0.5. To compare our method with basic and strong baseline methods, we conduct 5-way 1-shot and 5-way 5-shot classification tasks on the FSL test set, referred to as the novel set of meta-testing. For each FSL episode, we randomly select five categories (i.e., 5-way) from the test set and then, for the selected five categories, randomly choose one image (i.e., 1shot) or five images (i.e., 5-shot) to optimize the classifier head of the framework in our method and then evaluate the classification performance using 15 query images for each category. Finally, similar to previous FSL studies (Xie et al. 2022), we conduct 2000 episodes for each run and report the average results over 5 runs with 95% confidence interval. |