/Airbus_SDC

Code for the Kaggle competition "Airbus Ship Detection Challenge"

Primary LanguageJupyter Notebook

Airbus_SDC

Code for my first Kaggle competition "Airbus Ship Detection Challenge"

Implementation Details

In this competition I explored a host of new ideas:

  1. Importance Sampling and image weighting.
  2. Ship Cropping and image scaling.
  3. Dice, Brier, and Focal losses.
  4. U-Nets
  5. Global Convolutional Nets (GCN)
  6. Spatial and Channel Squeeze & Excitation (scSE) Blocks

Here are some of the results from my best (0.82) submission. The second column is the prediction and the third column is the ground truth.

Prediction compared to ground truth