/Indoors-Traversability-Estimation-with-Less-Labels-for-Mobile-Robots

Primary LanguagePythonMIT LicenseMIT

Indoors-Traversability-Estimation-with-Less-Labels for Mobile Robots

  1. Open a terminal window
  2. Clone the repo

Install the requirements on your machine

pip install -r requirements.txt

Fine-tuning on your dataset

For fine-tuning on your dataset using e.g. the fine tuned ViT:

  1. Open ViT_fine_tuned.py
  2. Specify the train and test dataset paths as '/home/../../set'
  3. Run python3 VIT_fine_tuned.py

You can follow the exact same process for fine-tuning with pretrained ResNet50

Using Ensemble GAN as described by Hirose et al. here

For training on a large dataset

  1. Arrange your folder structure as /../../data/class/
  2. Set the dataroot path '/../../data/'
  3. Run python3 GAN_ensemble_train.py

The model saves thre .pth files each corresponding to the models of the ensemble

For testing

  1. Open the script gan_classifier.py
  2. Specify the saved models to load as netD.load_state_dict(torch.load('model.pth'))
  3. Specify the train and test path as '/../../train/' and '/../../test/' 4.Run python3 gan_classifier.py

HERACLEiA dataset

Link to the dataset here