[ Discriminative Subgraphs for Discovering Family Photos ]

• We apply CORK algorithm for classifying group photos into family and non-family.

[ Requirements ]

• Python Libraries based on python 2.7.9
  • networkx==1.9.1
  • numpy==1.9.2
  • Pillow==2.9.0
  • requests==2.7.0
  • LibSVM==3.20: you may install manually ( https://www.csie.ntu.edu.tw/~cjlin/libsvm/#download )
• CORK algorithm: you can use one of them.
  • (original) by Marisa Thoma: http://www.dbs.ifi.lmu.de/~thoma/pub/sam2010/sam2010.zip
  • (extension) by Lei Zhao: https://code.google.com/p/grad-proj/source/browse/#svn%2Ftrunk%2Fgspancpp%2Fbin%253Fstate%253Dclosed

[ Querying Example ]

• First of all, you need to get a subsrciption key of face detection API from Project Oxford AI: https://www.projectoxford.ai/face

• it returns a json with age and gender and face positions in an image.

• Fill your subscription key into 'projection.py',

  request.add_header('Ocp-Apim-Subscription-Key', 'FILL YOUR SUBSCRIPTION KEY of OXFORD API')
  

• Then, execute querying.py

• If you want to test your own image, replace the q1.jpg with yours in 'query' folder.

• If you want to change the train model, chage the model number in querying.py,

  model_path = 'model_1'
  

[ CORK and create_trainmodel.py ]

• You can build your own train model using CORK algorithm and 'create_trainmodel.py'
• CORK returns the discriminative subgraphs such as 'Train_Cutoff4_CORKMAX_n2.subg' in 'train\model_1' folder
• 'create_trainmodel.py' tranform its type to some files: face graps, matrix of subgraphs
• Chen's dataset is in 'train\model_1\label' and our dataset is in 'train\model_1\label'
  • Both 2 dataset are rearranged from a public dataset: http://chenlab.ece.cornell.edu/people/Andy/ImagesOfGroups.html