leonardo-lyh / dn4

The Pytorch code of "Revisiting Local Descriptor based Image-to-Class Measure for Few-shot Learning", CVPR 2019.

DN4 in PyTorch

We provide a PyTorch implementation of DN4 for few-shot learning. If you use this code for your research, please cite:

Revisiting Local Descriptor based Image-to-Class Measure for Few-shot Learning.
Wenbin Li, Lei Wang, Jinglin Xu, Jing Huo, Yang Gao and Jiebo Luo. In CVPR 2019.

Prerequisites

• Linux
• Python 3
• Pytorch 0.4 or 1.0
• GPU + CUDA CuDNN
• pillow, torchvision, scipy, numpy

Getting Started

Installation

• Clone this repo:

git clone https://github.com/WenbinLee/DN4.git
cd DN4

• Install PyTorch 1.0 and other dependencies.

Datasets

• miniImageNet.
• StanfordDog.
• StanfordCar.
• CUB-200.
  Thanks Victor Garcia for providing the miniImageNet dataset. In our paper, we just used the CUB-200 dataset. In fact, there is a newer revision of this dataset with more images, see Caltech-UCSD Birds-200-2011. Note, if you use these datasets, please cite the corresponding papers.

miniImageNet Few-shot Classification

• Train a 5-way 1-shot model based on Conv64F or ResNet256F:

python DN4_Train_5way1shot.py --dataset_dir ./datasets/miniImageNet --data_name miniImageNet
or
python DN4_Train_5way1shot_Resnet.py --dataset_dir ./datasets/miniImageNet --data_name miniImageNet

• Test the model (specify the dataset_dir, basemodel, and data_name first):

python DN4_Test_5way1shot.py --resume ./results/DN4_miniImageNet_Conv64F_5Way_1Shot_K3/model_best.pth.tar --basemodel Conv64F
or
python DN4_Test_5way1shot.py --resume ./results/DN4_miniImageNet_ResNet256F_5Way_1Shot_K3/model_best.pth.tar --basemodel ResNet256F

• The results on the miniImageNet dataset (If you set neighbor_k as 1, you may get better results in some cases):

Fine-grained Few-shot Classification

• Data prepocessing (e.g., StanfordDog)
• Specify the path of the dataset and the saving path.
• Run the preprocessing script.

#!./dataset/StanfordDog_prepare_csv.py
python ./dataset/StanfordDog_prepare_csv.py

• Train a 5-way 1-shot model:

python DN4_Train_5way1shot.py --dataset_dir ./datasets/StanfordDog --data_name StanfordDog

• Test the model (specify the dataset_dir and data_name first):

python DN4_Test_5way1shot.py --resume ./results/DN4_StanfordDog_Conv64F_5_Way_1_Shot/model_best.pth.tar --basemodel Conv64F

• The results on the fine-grained datasets:

Citation

If you use this code for your research, please cite our paper.

@inproceedings{li2019DN4,
  title={Revisiting Local Descriptor based Image-to-Class Measure for Few-shot Learning},
  author={Li, Wenbin and Wang, Lei and Xu, Jinglin and Huo, Jing and Gao Yang and Luo, Jiebo},
  booktitle={CVPR},
  year={2019}
}