STAN-OSFGR-code

This is the code for our paper: "Hierarchical Attention Network for Open-Set Fine-Grained Image Recognition"[1], which has been accepted by T-CSVT. The code is based on https://github.com/sgvaze/osr_closed_set_all_you_need [2]

In addition, the Supplementary Material of the paper has been provided in "Supplementary Material of the Paper: Hierarchical Attention Network for Open-Set Fine-Grained Image Recognition.pdf".

Environments

The code is configured by PyTorch 1.7.1, torchvision 0.8.2, Python 3.7.7.

Datasets Preparation

Stanford-Cars

Please download the original Stanford-Cars datasets from https://ai.stanford.edu/~jkrause/cars/car_dataset.html, put the dataset in the "./dataset/stanford_cars" folder, and run the two steps in "process.py" in the subfolder "./dataset/stanford_cars/" for processing the format.

CUB

Please download the dataset from http://www.vision.caltech.edu/visipedia/CUB-200.html

Aircraft

Please download the dataset from https://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/

Trained Weights of the Proposed STAN-OSFGR

Stanford-Cars

Please download from https://drive.google.com/file/d/1vQ5BmvUskp8fFVRHbwB2jzg5azmFyVyg/view?usp=sharing

CUB

Please download from https://drive.google.com/file/d/1j1U-0nC0IMH6S_9VcD_CnF24bNVkE1ur/view?usp=sharing

Aircraft

Please download from https://drive.google.com/file/d/1OJrUfes5XBHdFyQyzf8AiJJ2NxlQRBe7/view?usp=sharing

Run

CUB/Aircraft

To train the model on the CUB/Aircraft datasets, run:

python train_CUB_Aircraft.py

To test the trained weights on the CUB datasets, run:

python test_CUB_Aircraft.py --dataset cub

To test the trained weights on the Aircraft datasets, run:

python test_CUB_Aircraft.py --dataset aircraft

Stanford-Cars

To train the model on the Stanford-Cars datasets, run:

python train_Stanford_Cars.py

To test the trained weights on the Stanford-Cars datasets, run:

python test_Stanford_Cars.py

[1] Jiayin Sun, Hong Wang, Qiulei Dong. Hierarchical Attention Network for Open-Set Fine-Grained Image Recognition[J]. IEEE Transactions on Circuits and Systems for Video Technology, DOI: 10.1109/TCSVT.2023.3325001, 2023.

[2] Vaze S, Han K, Vedaldi A, et al. Open-set recognition: A good closed-set classifier is all you need[C]. ICLR, 2022.