Compositional Prototypical Networks for Few-Shot Classification

Authors: Qiang Lyu, Weiqiang Wang

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate that our method can achieve state-of-the-art results on different datasets and settings.
Researcher Affiliation Academia School of Computer Science and Technology, University of Chinese Academy of Sciences
Pseudocode No The paper describes its method using textual explanations and equations, but it does not include any pseudocode or algorithm blocks.
Open Source Code Yes The code is available at https://github.com/fikry102/CPN.
Open Datasets Yes We conduct the experiments on two datasets with human-annotated attribute annotations: Caltech UCSD-Birds 200-2011 (CUB) (Wah et al. 2011), and SUN Attribute Database (SUN) (Patterson et al. 2014).
Dataset Splits Yes As (Chen et al. 2019a), we divide CUB into 100 training classes, 50 validation classes, and 50 testing classes. As (Huang et al. 2021; Ji et al. 2022), we divide SUN into 580 training classes, 65 validation classes, and 72 testing classes.
Hardware Specification No The paper mentions using 'Res Net12' and 'Conv4' as feature extractors, which are neural network architectures, not specific hardware. No details about GPUs, CPUs, or other hardware used for computation are provided.
Software Dependencies No The paper mentions software components like 'Res Net12', 'Conv4', 'SGD optimizer', and 'Dropblock regularization', but it does not specify any version numbers for these software dependencies (e.g., PyTorch or TensorFlow versions).
Experiment Setup Yes We use the SGD optimizer with a momentum of 0.9 and weight decay of 5 10 4. Following (Yang, Wang, and Chen 2022), we adopt the random crop, random horizontal flip and erasing, and color jittering to perform data augmentation. Dropblock (Ghiasi, Lin, and Le 2018) regularization is used to reduce overfitting. In the pre-training stage, we train the feature extractor for 30 epochs. In the meta-training stage, we train our model for 10 epochs.