This repository contains the code for the paper "Discriminative Suprasphere Embedding for Fine-Grained Visual Categorization"

• Install Torch with CUDA GPU
  Torch {http://torch.ch/docs/getting-started.html }
• Install cuDNN and the Torch cuDNN bindings
  cuDNN bindings {https://github.com/soumith/cudnn.torch/tree/R4 }
• Notes: For convenience, we recommend using docker

• Download CUB-200-2011 {http://www.vision.caltech.edu/visipedia/CUB-200-2011.html }
• Download pretrained ResNet-101 model {https://github.com/facebook/fb.resnet.torch/tree/master/pretrained }
• Arrange the dataset so that it contains a \train and a \val directory, which each contain sub-directories for every label. For example:

  "train/<label1>/<image.jpg>  
   train/<label2>/<image.jpg>  
   val/<label1>/<image.jpg>  
   val/<label2>/<image.jpg>"  

• To achieve this step you can use these files,

  \tmp\dataset\move.py    
  \tmp\dataset\train_images.txt    
  \tmp\dataset\test_images.txt    

• Place the rearranged dataset in tmp\dataset\CUB
• Place the pretrained ResNet-101 model in tmp\models

• Run the script inside script\script.txt
• Get the dataset index file tmp\dataset\cubsphere.t7
• Get the fine-tuned model tmp\result\cub_fine_tuned_model\model_best.t7
• Obtain classification accuracy

• Place the original downloaded dataset in visualization\data\CUB_200_2011
• Place the dataset index file tmp\dataset\cubsphere.t7 in visualization\data
• Place the fine-tuned model tmp\result\cub_fine_tuned_model\model_best.t7 in visualization\model
• Run the scripts under visualization\ in sequence according to the file name number
• Obtain Phase Activation Map (PAM), Class Contribution Map (CCM), Mean IoU, Discriminative localization results.