appcoreopc / kaggle_state_farm

VGG & Resnet Neural Networks for Kaggle's State Farm Distracted Driver Detection contest (Tensorflow)

Kaggle/State Farm Image Recognition Challenge

Given a dataset of 2D dashboard camera images, State Farm is challenging Kagglers to classify each driver's behavior. Are they driving attentively, wearing their seatbelt, or taking a selfie with their friends in the backseat?

Jupyter Notebook

Check Notebook for an overview of the implementation

Model Architectures

VGG Neural Network

• Developed by the Visual Geometry Group at the University of Oxford.
  
• It is class of very deep Convolutional Networks for large-scale Visual Recognition tasks.
  
• Won the first and the second places in the localisation and classification tasks respectively at the ImageNet ILSVRC-2014 contest.

Deep Residual Neural Network (Resnet)

• Developed by Microsoft Research.
  
• Won 1sth place in classification tasks at the ImageNet ILSVRC-2015 contest.
  
• The residual learning framework makes is easier to optimize much deeper networks, while maintaining a relativaly low complexity.
• Regular neural networks tend to decrease in accuracy at large depths, due to information degradation.
• Residual learning introduce skip connections, which allow information flow into the deeper layers and enable us to have deeper networks with better accuracy.