Seasons Of Code

Winning a Deep Learning challenge    • Mohd Safwan    • Arpit Aggarwal   

WnCC - Seasons of Code

Seasons of Code is a programme launched by WnCC along the lines of the Google Summer of Code. It provides one with an opprtunity to learn and participate in a variety of interesting projects under the mentorship of the very best in our institute.

List of Running Projects

Winning a Deep Learning challenge

Winning a Deep Learning challenge

The grand challenge website hosts multiple challenges on biomedical imaging. We are interested in getting good results and hopefully winning the Head Circumference challenge .

Anyone interested should read the problem statement thoroughly. We propose two solutions for the problem, running semantic segmentation and then fitting an ellipse on the mask, and regressing for the ellipse’s parameters using a Deep CNN. Our initial approach can be viewed here

This project will be heavy to say the least, but it will be rewarding. By the end we hope you will learn Deep Learning and will be able to appreciate its importance in our society. You’ll be able to understand and implement essential DL papers in PyTorch.

The following points must be included in the proposal for the project:

  • Your motivation and understanding of the project
  • Background in ML/DL (include your previous projects)
  • How do you want to approach the problem, you thoughts/remarks.
  • Experience with Python and scientific libraries

Tentative Timeline:

Week Work
Week 1: Finish Linear Algebra, Vector Calculus, and Statistics refresher
Week 2: Install Ubuntu, set up a development environment
Week 3: Learn/Brush-up Python, Torch, Jupyter, Numpy, Image Processing, Unix commands
Week 4: Learn Linear Regression, Logistic Regression, Neural Networks
Week 5: Read up on the use cases and building blocks of Deep Learning, data augmentation.
Week 6: Implement a fully connected network from scratch and train it.
Week 7: Learn PyTorch, implement a basic CNN for the same task using PyTorch.
Week 8: Get familiar with dataset of competition.
Week 9: Build a U-Net based segmentation model and a VGG based regression model.