nust-machine-intelligence-laboratory / crssc Goto Github PK

Datasets and source code for our paper CRSSC: Salvage Reusable Samples from Noisy Data for Robust Learning

After creating a virtual environment of python 3.5, run pip install -r requirements.txt to install all dependencies

The code is currently tested only on GPU.

• Data Preparation

  Download data into working directory and decompress them using

  wget https://web-fgvc-496-5089-sh.oss-cn-shanghai.aliyuncs.com/web-aircraft.tar.gz
  wget https://web-fgvc-496-5089-sh.oss-cn-shanghai.aliyuncs.com/web-bird.tar.gz
  wget https://web-fgvc-496-5089-sh.oss-cn-shanghai.aliyuncs.com/web-car.tar.gz
  tar -xvf web-aircraft.tar.gz
  tar -xvf web-bird.tar.gz
  tar -xvf web-car.tar.gz
  

• Source Code

  • If you want to train the whole model from beginning using the source code, please follow the subsequent steps.

    • Download dataset of web-aircraft/web-bird/web-car into the working directory as needed.
    • In train.sh
      • modify CUDA_VISIBLE_DEVICES to proper cuda device id.
      • modify DATA_BASE to the desired dataset and modify N_CLASSES accordingly.
      • modify NET to the desired backbone network.
    • Activate virtual environment (e.g. conda) and then run the script

      bash train.sh
      

• Demo

  • If you just want to do a quick test on the model and check the final recognition performance, please follow the subsequent steps.

    • Download one of the following trained models into model/ using

      wget https://fg-crssc-sh.oss-cn-shanghai.aliyuncs.com/web-aircraft_bcnn_best_epoch_76.4776.pth
      wget https://fg-crssc-sh.oss-cn-shanghai.aliyuncs.com/web-bird_bcnn_best_epoch_77.4249.pth
      wget https://fg-crssc-sh.oss-cn-shanghai.aliyuncs.com/web-car_bcnn_best_epoch_76.6447.pth
      

    • Activate virtual environment (e.g. conda)
    • In demo.sh
      • modify CUDA_VISIBLE_DEVICES to proper cuda device id.
      • modify DATA_BASE, N_CLASSES and NET according to the model downloaded.
    • Run demo using bash demo.sh

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

@inproceedings{sun2020crssc,
title={CRSSC: Salvage Reusable Samples from Noisy Data for Robust Learning},
author={Zeren Sun, Xian-Sheng Hua, Yazhou Yao, Xiu-Shen Wei, Guosheng Hu, Jian Zhang},
booktitle={ACM International Conference on Multimedia (ACM MM)},
year={2020}
}