Visualization

• Visualizing evidences for a deep network's classification
• CNN visualization
  

  

• CNN+LSTM (LRCN) visualization
• Original (ApplyEyeMakeup, UCF101 dataset)
  

  
  

  
  
• Visualized major evidences (8 frames per sequence for LRCN)
  

  
  

  
  

  
  
• Original ("time" in Korean Sign Language, self-generated dataset)

  

• Visualized major evidences (8 frames per sequence for LRCN)

  

  
  

  
  
• Original ("how-many" in Korean Sign Language, self-generated dataset)

  
  

  

  
  

• Visualized major evidences (8 frames per sequence for LRCN)