DL-Project

by NINAD MHALGI and ABICHAL GHOSH

The file "DL_Midsem_U2Net.ipynb" contains the implementation of UNet, whereas "Final_DL_Compre.ipynb" contains the implementation of U2Net Compre Final

The ipynb notebook contains one of the possible training and test combination which is

  trained - ECSSD, validated - ECSSD

However, for the comparitive study, the following combinations have been used

  trained - ECSSD, validated - ECSSD
  trained - ECSSD, validated - MSRA
  trained - ECSSD, validated - DUTS
  trained - MSRA, validated - ECSSD
  trained - MSRA, validated - MSRA
  trained - MSRA, validated - DUTS

In order to obatain all the combination results, the datasets and models have been provided in the form of google drive link

If one wants to train a new model on a new dataset follow the steps -

  1 download the dataset zip file
  2 unzip the file
  3 set the train image and mask pathway to the intended
  4 run

If one wants to validate a pretrained model, follow the steps -

  1 skip to validation
  2 download the intented pretrained model file from google drive link
  3 load the pretrained model
  4 run

If one wants to validate a model on a different test/validation set, follow the steps -

  1 download the validation dataset zip file
  2 unzip the file
  3 skip to the validation image and mask path declaration
  4 change the declaration to intented
  5 run