Data-driven Meta-set Based Fine-Grained Visual Recognition

Introduction

This is the source code for our paper Data-driven Meta-set Based Fine-Grained Visual Recognition.

Network Architecture

The architecture of our proposed peer-learning model is as follows

Installation

After creating a virtual environment of python 3.7, run pip install -r requirements.txt to install all dependencies

How to use

The code is currently tested only on GPU

• Data Preparation

  • Download web data into project root directory and uncompress them using

    wget https://wsnfg-sh.oss-cn-shanghai.aliyuncs.com/web-bird.tar.gz
    wget https://wsnfg-sh.oss-cn-shanghai.aliyuncs.com/web-car.tar.gz
    wget https://wsnfg-sh.oss-cn-shanghai.aliyuncs.com/web-aircraft.tar.gz
    tar -xvf web-bird.tar.gz
    tar -xvf web-car.tar.gz
    tar -xvf aircraft-car.tar.gz
    

  • If you already have the benchmark datasets, you can directly use part of them as the validation sets. Otherwise, download validation data into project root directory and uncompress them using

    wget https://dmfgr.oss-cn-hongkong.aliyuncs.com/validation_data.zip
    unzip validation_data.zip
    

• Source Code

  • If you want to train the whole network from begining using source code on the web fine-grained dataset, please follow subsequent steps

    • Modify CUDA_VISIBLE_DEVICES to proper cuda device id and data_base, val_base to proper dataset in train.sh. You can directly utilize the benchmark dataset as the validation set by changing meta_number parameter in the train.py. meta_number controls the number of images per category in the validation set, and it is set to be 10 by defaut.

    • Activate virtual environment(e.g. conda) and then run the scriptbash train.sh to train the model. We recommand you to use resnet-18 backbone, because it needs less memory and time to train.

  

Citation

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

@inproceedings{zhang2020data,
title={Data-driven Meta-set Based Fine-Grained Visual Recognition},
author={Chuanyi Zhang, Yazhou Yao, Xiangbo Shu, Zechao Li, Zhenmin Tang, Qi Wu},
booktitle={ACM International Conference on Multimedia (ACM MM)},
year={2020}
}