For the uninitiated, image analysis can be daunting, especially when it comes to deep learning. In this talk, we'll use Kaggle's 2018 Data Science Bowl as a case study in image analysis. Even if you've never worked with an image array before, you'll walk away with some foundational tools for image analysis and an understanding of how some common deep learning models perform so well.
This talk was last given to the Nashville Data Science MeetUp group on April 24, 2018.
Stephen Bailey is currently an NIH Fellow at Vanderbilt University, using magnetic resonance imaging to understand how education changes brain structure and function. He placed in the 12th percentile for the 2018 Data Science Bowl.
- slides.pptx, slides.pdf: Slides from the 4/24 Data Science MeetUp talk.
- dsb-u-net-starter.ipynb: Jupyter Notebook from Kjetil Amdal-Sevik's Kaggle kernel.
- dsb-teaching-notebook.ipynb: Jupyter Notebook from Stephen Bailey's Kaggle kernel.
- Coursera’s Deep Learning specialization from deeplearning.ai.
- “U-Net: Convolutional Networks for Biomedical Image Segmentation” by Ronneberger, Fischer and Brox
- Kaggle’s Data Science Bowl 2018
- 1st Place Solution from ods.ai
- 4th Place U-Net + Watershed from Nuclear Vision
- 11th Place Open Solution from Zheng Li