Masked Feature Generation Network for Few-Shot Learning

Authors: Yunlong Yu, Dingyi Zhang, Zhong Ji

IJCAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In extensive experiments on four FSL benchmarks, MFGN performs competitively and outperforms the state-of-the-art competitors on most of the few-shot classification tasks.
Researcher Affiliation Academia Yunlong Yu1 , Dingyi Zhang1 , Zhong Ji2 1 College of Information Science and Electronic Engineering, Zhejiang University 2 School of Electrical and Information Engineering, Tianjin University {yuyunlong, dyzhang}@zju.edu.cn, jizhong@tju.edu.cn
Pseudocode Yes Algorithm 1 Pseudocode of MFGN in a Py Torch-like style.
Open Source Code No The paper does not contain any explicit statements or links indicating that the source code for the methodology is openly available.
Open Datasets Yes We have conducted experiments on four datasets that are Caltech UCSD Birds (CUB) [Wah et al., 2011], Stanford Dogs [Khosla et al., 2011], mini Image Net [Vinyals et al., 2016], and CIFAR-FS [Bertinetto et al., 2018].
Dataset Splits Yes Following the split proposed in [Chen et al., 2018], we divide the data into three disjoint sets, 100, 50, and 50 categories, respectively for training, validation, and test. ... Following the split proposed in [Ravi and Larochelle, 2016], we divide the dataset with 64, 16, and 20 classes for training, validation, and test, respectively. The CIFAR-FS is a few-shot classification dataset built on the CIFAR100. Following [Bertinetto et al., 2018], we use 64/16/20 classes for training/validation/test, respectively.
Hardware Specification No The paper mentions using a CNN backbone and a transformer, but does not provide any specific details about the hardware (e.g., GPU model, CPU type, memory) used for running the experiments.
Software Dependencies No The paper mentions 'Py Torch-like style' pseudocode but does not list specific software dependencies with version numbers, such as PyTorch version, Python version, or other libraries.
Experiment Setup Yes To train the model, we apply the SGD optimizer with a momentum of 0.9 and a weight decay of 5e 4. The batch size is set to 32. For the fine-grained datasets, we train the models with 80 epochs. The learning rate is initialized as 0.05 and decayed with a factor of 0.1 at 60 and 70 epochs, respectively. For the coarse-grained datasets, we train the models with 60 epochs. The learning rate is initialized as 0.05 and decayed with a factor of 0.1 at 40 and 50 epochs, respectively. ... If not specific, both λ and γ are set to 1; the number of synthesized visual features is set to 5 for all datasets.