This is the source code for our paper Co-LDL: A Co-training-based Label Distribution Learning Method for Tackling Label Noise

After creating a virtual environment of python 3.6, run pip install -r requirements.txt to install all dependencies

The code is currently tested only on GPU.

• Data preparation

  Create a folder Datasets in the project root directory and download data into this folder using

  wget https://web-fgvc-496-5089-sh.oss-cn-shanghai.aliyuncs.com/web-aircraft.tar.gz
  tar -xvf web-aircraft.tar.gz
  
  # optional
  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-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 subsequent steps:
    • Prepare data
    • Modify GPU in the run.sh
    • Activate virtual environment (e.g. conda) and then run

      bash run.sh
      

• Demo

  • If you just want to reproduce our result, please follow subsequent steps:
    • Prepare data
    • Download the following trained model

      wget https://webly-coldl.oss-cn-shanghai.aliyuncs.com/aircraft.pth
      

    • Activate virtual environment (e.g. conda) and then run

      python demo.py --arch resnet50 --model_path aircraft.pth --dataset web-aircraft --nclasses 100 --gpu 4