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 imageslib
- contains common functions / classes / variables used in different solutionsmerge_submissions
- solution, which creates new submision merging several existing ones with specified weightsunet_with_cropping
- main solution which uses U-Net neural network for image segmentationvisualization
- contains IPython notebook to visualize predicted masks
Model weights (full images): carvana_unet.pth
Model weights (crops): carvana_unet_crop.pth