About baseline

AP-CNN

Code release for Weakly Supervised Attention Pyramid Convolutional Neural Network for Fine-Grained Visual Classification (TIP2021).

Dependencies

Python 3.6 with all of the pip install -r requirements.txt packages including:

torch == 0.4.1
opencv-python
visdom

Data

Download the FGVC image data. Extract them to data/cars/, data/birds/ and data/airs/, respectively.

Stanford-Cars (cars)

  -/cars/
     └─── car_ims
             └─── 00001.jpg
             └─── 00002.jpg
             └─── ...
     └─── cars_annos.mat

CUB-200-2011 (birds)

  -/birds/
     └─── images.txt
     └─── image_class_labels.txt
     └─── train_test_split.txt
     └─── images
             └─── 001.Black_footed_Albatross
                       └─── Black_Footed_Albatross_0001_796111.jpg
                       └─── ...
             └─── 002.Laysan_Albatross
             └─── ...

FGVC-Aircraft (airs)

   -/airs/
     └─── images
             └─── 0034309.jpg
             └─── 0034958.jpg
             └─── ...
     └─── variants.txt
     └─── images_variant_trainval.txt
     └─── images_variant_test.txt

Preprocess images.

For birds: python utils/split_dataset/birds_dataset.py
For cars: python utils/split_dataset/cars_dataset.py
For airs: python utils/split_dataset/airs_dataset.py

Training

Start:

python train.py --dataset {cars,airs,birds} --model {resnet50,vgg19} [options: --visualize] to start training.

For example, to train ResNet50 on Stanford-Cars: python train.py --dataset cars --model resnet50
Run python train.py --help to see full input arguments.

Visualize:

python -m visdom.server to start visdom server.
Visualize online attention masks and ROIs on http://localhost:8097.

Pretrained Checkpoints

Pretrained checkpoints with following settings are available on download link, with access code "kjqu".

Dataset	base model	accuracy(%)
CUB-200-2011	resnet50	88.4
Stanford-Cars	resnet50	95.3
FGVC-Aircraft	resnet50	94.0

Citation

If you find this paper useful in your research, please consider citing:

@ARTICLE{9350209,
author={Y. {Ding} and Z. {Ma} and S. {Wen} and J. {Xie} and D. {Chang} and Z. {Si} and M. {Wu} and H. {Ling}},
journal={IEEE Transactions on Image Processing},
title={AP-CNN: Weakly Supervised Attention Pyramid Convolutional Neural Network for Fine-Grained Visual Classification},
year={2021},
volume={30},
number={},
pages={2826-2836},
doi={10.1109/TIP.2021.3055617}}

Contact

Thanks for your attention! If you have any suggestion or question, you can leave a message here or contact us directly:

pris-cv / ap-cnn_pytorch-master Goto Github PK

ap-cnn_pytorch-master's Introduction

AP-CNN

Dependencies

Data

Training

Pretrained Checkpoints

Citation

Contact

ap-cnn_pytorch-master's People

Contributors

Stargazers

Watchers

Forkers

ap-cnn_pytorch-master's Issues

Recommend Projects

Recommend Topics

Recommend Org