/kaggle-carvana-challenge

Solution for Kaggle Carvana Image Masking Challenge

Primary LanguageJupyter Notebook

kaggle-carvana-challenge

Project contains solution for Kaggle Carvana Image Masking Challenge.

Project structure

  • avg_mask - baseline solution which applies average masks computed on train images to all test images
  • lib - contains common functions / classes / variables used in different solutions
  • merge_submissions - solution, which creates new submision merging several existing ones with specified weights
  • unet_with_cropping - main solution which uses U-Net neural network for image segmentation
  • visualization - contains IPython notebook to visualize predicted masks

Model weights (full images): carvana_unet.pth
Model weights (crops): carvana_unet_crop.pth