CPSN-OSFG.PyTorch

A simple PyTorch implementation of CPSN (Coupled Patch Similarity Network) for OSFG (One-Shot Fine-Grained Image Recognition).

Coupled Patch Similarity Network For One-Shot Fine-Grained Image Recognition, ICIP2021

Environment

This code is implemented on PyTorch and the experiments were done on a NVIDIA TITAN RTX GPU. So you may need to install

• Python==3.x
• torch==1.1.0 or above
• torchvision==0.3.0
• tqdm

Datasets

• CUB-200-2011.
  
• StanfordCar.
  In our paper, we just used the CUB-200-2011 dataset and StanfordCar dataset. Note, if you use these datasets, please cite the corresponding papers.

Usage

This repo contains CPSN with feature extractors using Conv64 / ResNet12 in PyTorch form, see ./models/.

• resnet: Conv64|ResNet12.
• nExemplars: number of training examples per novel category.
• nKnovel: number of novel categories.
• gpu_devices: gpu id.
• scale_cls: scale the metric scores.
• phase: use test or val dataset to early stop.
• num_classes: number of all training categories.

Train a 5-way 1-shot model based on Conv64 or ResNet12:

python ./train.py --dataset CUB  --resnet 0 --num_classes 100 --nExemplars 1
python ./train.py --dataset CUB  --resnet 1 --num_classes 100 --nExemplars 1

Test the model:

python ./test.py --dataset CUB  --resnet 0 --num_classes 100 --nExemplars 1
python ./test.py --dataset CUB  --resnet 1 --num_classes 100 --nExemplars 1

Citation

If this work is useful in your research, please cite

@inproceedings{tian2021coupled,
  title={Coupled Patch Similarity Network FOR One-Shot Fine-Grained Image Recognition},
  author={Tian, Sheng and Tang, Hao and Dai, Longquan},
  booktitle={2021 IEEE International Conference on Image Processing (ICIP)},
  pages={2478--2482},
  year={2021},
  organization={IEEE}
}

References

This implementation builds upon several open-source codes. Specifically, we have modified and integrated the following codes into this repository:

• CAN