Distribution Consistency Based Covariance Metric Networks for Few-Shot Learning
Authors: Wenbin Li, Jinglin Xu, Jing Huo, Lei Wang, Yang Gao, Jiebo Luo8642-8649
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments in two tasks, generic few-shot image classification and fine-grained fewshot image classification, demonstrate the superiority of the proposed Cova MNet. |
| Researcher Affiliation | Academia | 1National Key Laboratory for Novel Software Technology, Nanjing University, China 2Northwestern Polytechnical University, China 3University of Wollongong, Australia 4University of Rochester, USA |
| Pseudocode | No | The paper does not contain any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The source code can be available from https://github.com/Wenbin Lee/Cova MNet.git. |
| Open Datasets | Yes | The mini Image Net dataset was originally proposed by (Vinyals et al. 2016), a mini-version of Image Net derived from the ILSVRC-12 dataset (Russakovsky et al. 2015). There are 100 categories with 600 images per category in this dataset and the image resolution is 84 84. In this work, we follow the splits of this dataset used in (Ravi and Larochelle 2017), where 64, 16 and 20 categories are for training (auxiliary), validation and testing, respectively. |
| Dataset Splits | Yes | In this work, we follow the splits of this dataset used in (Ravi and Larochelle 2017), where 64, 16 and 20 categories are for training (auxiliary), validation and testing, respectively. Experimental Setting Typically, the 5-way 1-shot and the 5-way 5-shot classification tasks are conducted on this dataset. During the process of training, we employ the episodic training mechanism to learn the proposed Cova MNet model. There are totally 300, 000 episodes, each of which is constructed by a support set and a query set. For the 5-way 1-shot classification, there are 5 categories with 15 query images and 1 support image per category, i.e., 5 15+5 1 = 80 images in each episode. Similarly, for the 5-way 5-shot classification, there are 5 15 + 5 5 = 100 images in each episode. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used, such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | The paper mentions using the 'Adam algorithm (Kingma and Ba 2015)' but does not specify any software libraries or dependencies with version numbers (e.g., TensorFlow version, PyTorch version, Python version). |
| Experiment Setup | Yes | During the process of training, we employ the episodic training mechanism to learn the proposed Cova MNet model. There are totally 300, 000 episodes, each of which is constructed by a support set and a query set. For the 5-way 1-shot classification, there are 5 categories with 15 query images and 1 support image per category, i.e., 5 15+5 1 = 80 images in each episode. Similarly, for the 5-way 5-shot classification, there are 5 15 + 5 5 = 100 images in each episode. Besides, we adopt Adam algorithm (Kingma and Ba 2015) with an initial learning rate of 5 10 3 to optimize our Cova MNet model, where the learning rate is reduced by half for every 100, 000 episodes. |