Skip to content

Latest commit

 

History

History

scripts

README.md

Scripts to run all the major experiments

The above .sh files are used to run the individual experiments. Each .sh file has the hyperparameters used for running the specific experiment. The only modification required while executing these scripts is to change the path to the data.

Scripts for Training on Kaggle Data Only (Effect of Pretrained Weights for Initialization)

  • run_unet_ethz.sh - This script trains the UNet on Kaggle Data.
  • run_xception_ethz.sh - This script trains the X-UNet on Kaggle Data with pretrained initialization for the encoder.
  • run_xception_ethz_nopt.sh - This script trains the X-UNet on Kaggle Data with random initialization for the encoder.
  • run_resnet_ethz.sh - This script trains the ResUNet on Kaggle Data with pretrained initialization for the encoder.
  • run_resnet_ethz_nopt.sh - This script trains the ResUNet on Kaggle Data with random initialization for the encoder.

Scripts for Finetuning Experiments

  • run_unet_gmap.sh - This script trains the UNet on Google Data.
  • run_xception_gmap_nopt.sh - This script trains the X-UNet on Google Data with random initialization for the encoder.
  • run_xception_gmap.sh - This script trains the X-UNet on Google Data with pretrained initialization for the encoder.
  • run_resnet_gmap.sh - This script trains the ResUNet on Google Data with pretrained initialization for the encoder.
  • run_resnet_gmap_nopt.sh - This script trains the ResUNet on Google Data with random initialization for the encoder.
  • ft_unet_gmap.sh - This script finetunes the UNet on Kaggle Data. The model loaded for finetuning was trained on Google Data while being initialized randomly.
  • ft_xception_gmap_nopt.sh - This script finetunes the X-UNet on Kaggle Data. The model loaded for finetuning was trained on Google Data while being initialized randomly.
  • ft_xception_gmap_pt.sh - This script finetunes the X-UNet on Kaggle Data. The model loaded for finetuning was trained on Google Data while being initialized to pretrained weights.
  • ft_resnet_gmap_pt.sh - This script finetunes the ResUNet on Kaggle Data. The model loaded for finetuning was trained on Google Data with encoder being initialized to pretrained weights.
  • ft_resnet_gmap_nopt.sh - This script finetunes the ResUNet on Kaggle Data. The model loaded for finetuning was trained on Google Data with encoder being initialized randomly.

The models obtained from all the experiments and the ones used for finetuning are available here.