Image-Classification

• Use Deep Learning models from scratch like (Xception, DenseNet, AlexNet, AlexNet with Transformer)

• Apply Data Preparation steps :

  • Image Preprocessing :

    • Convert image from BGR to RGB
    • Resizing Images
    • Normalize Image by Dividing by Maximum

  • Data Augmentation For Train Like:

    • Horizontal flip
    • Vertical flip
    • Rotation range by 40
    • Width shift range
    • Height shift range
    • Zoom range
    • Fill mode by Nearest

• Apply OneHotEncoder to convert Classes to labels

• Split into train and validation

• Transfomer :

  • It uses a conventional CNN backbone to learn a 2D representation of an input image.
  • The model flattens it and supplements it with a positional encoding before passing it into a transformer encoder.
  • After features + positional encoding , additionally attends to the encoder.
  • After Encoder pass to linear layer then to classification layer ‘softmax’ .

• submit results on Kaggle competition and rank the best accuracy achieved in the competition.

Collaborators:

• Hager Khaled
  
• Nourhan Mohamed
  
• Nada Keshka
  
• Yasmin Hamada
  
• Fatma Salah
  
• Fatma Mahmoud