DeepLabV3-Road-Boundary-Estimation

To run a file individually, download or copy that file, upload/paste in Google Colab, Kaggle or Jupyter Notebook. Update IMG_PATH, MASK_PATH, IMG_SUB_PATH, MASK_SUB_PATH, ONEHOT_MASK according to your file structure.

To use Combined_’models_with_loss_function_and_weight_options’ download or copy that file, upload/paste in Google Colab, Kaggle or Jupyter Notebook.

1. Select the model you want to use as backbone in ‘model_name’
2. Select the loss function you want to use for model training in ‘loss_fucntion’
3. Select ‘Yes’ or ‘No’ depending on whether you want to handle class imbalance or not in ‘add_weight’ Then update IMG_PATH, MASK_PATH, IMG_SUB_PATH, MASK_SUB_PATH, ONEHOT_MASK according to your file structure.

If the folder structure of the dataset is like this: D:.
|---final dataset
|  |---mask
|  |  |---mask
|  |---test Image
|  |---test one hot
|  |---train Image
|  |  |---Image
|  |---train One hot
|  |  |---train One hot

Then the path veriables should be: IMG_PATH = /final dataset/train Image/
MASK_PATH = /final dataset/mask/
IMG_SUB_PATH = /final dataset/train Image/Image/      #training image
MASK_SUB_PATH = /final dataset/train One hot/One hot/   #Actually segment label
ONEHOT_MASK = /final dataset/mask/mask         #Here segmented label will be stored after preprocessing

This paper might be helpful for understanding this repository.

If you find this repository and paper helpful, we would appreciate using the following citations:

@article{das2021estimation,
  title={Estimation of road boundary for intelligent vehicles based on deepLabV3+ architecture},
  author={Das, Sunanda and Fime, Awal Ahmed and Siddique, Nazmul and Hashem, MMA},
  journal={IEEE Access},
  volume={9},
  pages={121060--121075},
  year={2021},
  publisher={IEEE}
}